Atlas
Open-source desktop app for Mac, Windows and Linux. Agents run on your computer, on your own server or in the vendor's cloud.
Product websitetryatlas.cc Compare with others Missing or wrong data?
- Desktop app
- Mac, Windows and Linux
- Browser
- Yes
- Phone
- Phone browser
- Command line
- Yes
- Agents
- Claude Code, Codex, Cursor Agent, OpenCode, Pi and Antigravity
- Your subscription
- Works with your own plan
- Agents run
- On your computer, on your own server and in the vendor's cloud
- Laptop closed
- No
- Isolation
- Worktree
- Parallel agents
- Yes
- Scheduled runs
- Yes
- Coordinator agent
- Yes
- Built-in issues
- No
- Pull requests
- Yes
- Review screen
- Yes
- Team
- Shared sessions and roles
- Self-hosting
- Yes
- Licence
- Apache-2.0
- Price from
- Free
- How it charges
- Free plan, subscription and bring your own key
- Paid plans from
- $11–20/month
Receipts
- Checked
- Findings
- 50 published · 48 cross-checked by a second agent · read them all
- Sources
- 58 public pages on 4 sites: tryatlas.cc, docs.tryatlas.cc, api.github.com, github.com
- Screenshot
- docs.tryatlas.cc, retrieved · first party docs · shows appearance, not capability
- Data
- /products/atlas.json · .md
- Method
- Same rules for all 103 researched ADEs. How we check · Report wrong data
What it does
Where you use it
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
Source and scope for Where you use itAgents it runs
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
Source and scope for Agents it runsAccounts and subscriptions
External agents run on the user’s own subscriptions or API keys; provider keys also drive timeline Q&A and memory extraction; paid plans add included AI ($8/$40/$80) via granted AI access; local embedding model only; no multi-account switching documented.
Source and scope for Accounts and subscriptionsWhere agents run
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
Source and scope for Where agents runHow work is kept apart
The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
Source and scope for How work is kept apartParallel work
Multiple sessions run concurrently (second session beside the first; threads at once; watched PRs keep sessions working in background) with per-thread concurrent-run quotas on paid plans; sidebar and timeline show running/waiting/settled/live states. No observed concurrency fixture was run.
Source and scope for Parallel workScheduled runs
No schedules or recurring tasks. Watched pull requests poll every 2 minutes and event-message the agent on check fail/pass, comments, reviews, and conflicts; incoming chat webhooks post external events to channels. Desktop must stay running (window may hide).
Source and scope for Scheduled runsGit and pull requests
Full GitHub PR workflow inside Atlas: branch/worktree checkout, Create PR, review threads, checks, rebase/merge conflict resolution, revert, branch delete — all over gh with Atlas never pushing for the user.
Source and scope for Git and pull requestsLicence and source
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
Source and scope for Licence and sourcePrice
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
Source and scope for PriceTyped filters
Every value links to its receipt below. Read plan and surface restrictions before using a filter to decide.
| Filter | Value |
|---|---|
| Desktop OS: macOS | Yes cross-checked |
| Desktop OS: Windows | Yes cross-checked |
| Desktop OS: Linux | Yes cross-checked |
| Web client | Yes cross-checked |
| Phone access: Native iOS | No cross-checked |
| Phone access: Native Android | No cross-checked |
| Phone access: Phone browser | Yes cross-checked |
| CLI | Yes cross-checked |
| Coding agents: Claude Code | Yes cross-checked |
| Coding agents: Codex | Yes cross-checked |
| Coding agents: Cursor Agent | Yes cross-checked |
| Coding agents: Gemini CLI | Unknown |
| Coding agents: OpenCode | Yes cross-checked |
| Coding agents: Pi | Yes cross-checked |
| Coding agents: Grok | Unknown |
| Coding agents: Aider | Unknown |
| Coding agents: Antigravity | Yes cross-checked |
| Use my subscription | Yes cross-checked |
| Where agents run: On your computer | Yes cross-checked |
| Where agents run: On your own server | Yes cross-checked |
| Where agents run: In the vendor's cloud | Yes cross-checked |
| Workspace isolation: Git worktree | Yes |
| Workspace isolation: Container | No |
| Workspace isolation: VM or microVM | No |
| Parallel runs | Yes cross-checked |
| Coordinator | Yes |
| Native issue tracking | No cross-checked |
| Unattended triggers | Yes cross-checked |
| Shared prompting | Yes cross-checked |
| Team roles | Yes cross-checked |
| SSO | Unknown |
| Pull request workflow | Yes cross-checked |
| Review interface | Yes cross-checked |
| Product license: Apache-2.0 | Yes. Apache-2.0 cross-checked |
| Product license: MIT | No. Apache-2.0 cross-checked |
| Product license: GPL-3.0 | No. Apache-2.0 cross-checked |
| Product license: GPL-3.0-or-later | No. Apache-2.0 cross-checked |
| Product license: AGPL-3.0 | No. Apache-2.0 cross-checked |
| Product license: AGPL-3.0-or-later | No. Apache-2.0 cross-checked |
| Product license: Elastic License 2.0 | No. Apache-2.0 cross-checked |
| Product license: Custom terms | No. Apache-2.0 cross-checked |
| Product license: Proprietary | No. Apache-2.0 cross-checked |
| Self-hosting | Yes cross-checked |
| Telemetry opt-out | Yes cross-checked |
| Free access | Yes cross-checked |
| Starting price: Free | Yes. Free cross-checked |
| Starting price: Up to $20/month | No. Free cross-checked |
| Starting price: $21–40/month | No. Free cross-checked |
| Starting price: $41–100/month | No. Free cross-checked |
| Starting price: Over $100/month | No. Free cross-checked |
| Starting price: Contact for pricing | No. Free cross-checked |
| Pricing model: Free, no paid plan | No cross-checked |
| Pricing model: Free plan | Yes cross-checked |
| Pricing model: Subscription | Yes cross-checked |
| Pricing model: Per seat | No |
| Pricing model: Usage-based | No |
| Pricing model: Per machine or compute | No |
| Pricing model: Bring your own key | Yes cross-checked |
| Pricing model: One-time purchase | No |
| Paid plans from: No paid plan | No. $11–20/month |
| Paid plans from: $10 or less/month | No. $11–20/month |
| Paid plans from: $11–20/month | Yes. $11–20/month |
| Paid plans from: $21–40/month | No. $11–20/month |
| Paid plans from: $41–100/month | No. $11–20/month |
| Paid plans from: Over $100/month | No. $11–20/month |
| Paid plans from: Contact sales | No. $11–20/month |
| Lifecycle status: Active | Unknown |
| Lifecycle status: Sunsetting | Unknown |
| Keeps going with laptop closed | No cross-checked |
| Open-source license | Yes. Apache-2.0 cross-checked |
| Any desktop app | Yes cross-checked |
| Paid plan up to $20/month | Yes. $11–20/month |
| Paid plan up to $40/month | Yes. $11–20/month |
| Any phone client | Yes cross-checked |
Filter receipts
Desktop OS: macOS: Yes cross-checked
Desktop OS: macOS
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Desktop os: macOS 11+ Apple silicon/Intel, Windows 10+ x64, Linux x86-64 packages (deb/rpm/Arch/AppImage/tarball; WebKitGTK 4.1).
- Cli: atlas-cli on macOS and Linux for capture/sync/memory (not on Windows); desktop also installs an `atlas` folder-opener on macOS and `atl`/`tryatlas` on Linux.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
Desktop OS: Windows: Yes cross-checked
Desktop OS: Windows
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Desktop os: macOS 11+ Apple silicon/Intel, Windows 10+ x64, Linux x86-64 packages (deb/rpm/Arch/AppImage/tarball; WebKitGTK 4.1).
- Cli: atlas-cli on macOS and Linux for capture/sync/memory (not on Windows); desktop also installs an `atlas` folder-opener on macOS and `atl`/`tryatlas` on Linux.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
Desktop OS: Linux: Yes cross-checked
Desktop OS: Linux
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Desktop os: macOS 11+ Apple silicon/Intel, Windows 10+ x64, Linux x86-64 packages (deb/rpm/Arch/AppImage/tarball; WebKitGTK 4.1).
- Cli: atlas-cli on macOS and Linux for capture/sync/memory (not on Windows); desktop also installs an `atlas` folder-opener on macOS and `atl`/`tryatlas` on Linux.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
Web client: Yes cross-checked
Web client
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Browser: Released web app at app.tryatlas.cc: inbox, timeline, projects, chat, ROI, settings, share links and live remote-control rooms with prompt/stop/model controls. Runs no coding agents.
- Notifications: Native OS banners + in-app center (agent finish/question/permission, mentions, long terminal commands), per-kind switches, email notifications with quiet hours.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
Phone access: Native iOS: No cross-checked
Phone access: Native iOS
Typed value: No.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Mobile apps: No native phone app: mobile app is in progress, not in the App Store or Google Play yet, and would not run agents anyway.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: "In progress: being built, and not in a release yet" and "not in the App Store or Google Play yet." Planned iOS/Android client for chat, timeline, inbox, projects, members, dashboard; no agent execution; no calls (join from web/desktop); no push notifications; no Spaces/drafts/shared threads. Device-code sign-in like desktop.
Phone access: Native Android: No cross-checked
Phone access: Native Android
Typed value: No.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Mobile apps: No native phone app: mobile app is in progress, not in the App Store or Google Play yet, and would not run agents anyway.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: "In progress: being built, and not in a release yet" and "not in the App Store or Google Play yet." Planned iOS/Android client for chat, timeline, inbox, projects, members, dashboard; no agent execution; no calls (join from web/desktop); no push notifications; no Spaces/drafts/shared threads. Device-code sign-in like desktop.
Phone access: Phone browser: Yes cross-checked
Phone access: Phone browser
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Browser: Released web app at app.tryatlas.cc: inbox, timeline, projects, chat, ROI, settings, share links and live remote-control rooms with prompt/stop/model controls. Runs no coding agents.
- Mobile browser: Remote-control room page is built for a phone screen and carries the live session with prompt/stop/model controls; no native app needed for it.
- Task actions: Desktop starts/steers sessions (second session beside it, worktree starts, settle/snooze), approvals via permission prompts, review in diffs/PR tabs, resume repaints transcript from disk; web rooms prompt/stop/switch model.
- Cross device continuity: Remote-control rooms and org sync continue the same session from another computer or phone browser, but the agent keeps running on the desktop; "pick up work on any device" is roadmapped.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
CLI: Yes cross-checked
CLI
Typed value: Yes.
atlas-cli covers account/org/setup/agents/capture/sync/memory/status/update; desktop installs a folder-opener; incoming chat webhooks and a memory MCP server/bridge exist; no documented public REST API operations, SDK, or outgoing webhooks; personal access token for scripts exists without documented operations.
- Cli: atlas-cli: login/org/init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync serve, memory browse/export/import/verify, update/uninstall; piped plain-text output with flags.
- Mcp server: Atlas exposes a memory tool server to sessions (HTTP MCP) plus atlas-cli mcp-bridge; session create/steer/stop operations over MCP are not documented.
- Operations: CLI inspects (status, sessions list) and manages capture/memory; no steer/stop/review operations documented.
- No public REST API operations, SDK, or outgoing webhooks documented.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
Coding agents: Claude Code: Yes cross-checked
Coding agents: Claude Code
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed harness operation: Vendor shows Claude Code, Codex and ACP agents side by side on the same codebase with mid-task switching; no two-harness observed fixture was run.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
Coding agents: Codex: Yes cross-checked
Coding agents: Codex
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed harness operation: Vendor shows Claude Code, Codex and ACP agents side by side on the same codebase with mid-task switching; no two-harness observed fixture was run.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
Coding agents: Cursor Agent: Yes cross-checked
Coding agents: Cursor Agent
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed harness operation: Vendor shows Claude Code, Codex and ACP agents side by side on the same codebase with mid-task switching; no two-harness observed fixture was run.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
Coding agents: OpenCode: Yes cross-checked
Coding agents: OpenCode
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed harness operation: Vendor shows Claude Code, Codex and ACP agents side by side on the same codebase with mid-task switching; no two-harness observed fixture was run.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
Coding agents: Pi: Yes cross-checked
Coding agents: Pi
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
Coding agents: Antigravity: Yes cross-checked
Coding agents: Antigravity
Typed value: Yes.
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Harnesses: Atlas Agent plus registry installs of Claude Code, Codex, Cursor, OpenCode and more; CLI hooks Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi.
- Session integration: Atlas runs external agents through ACP and keeps each chat history itself (resume/import across agents); CLI hooks agents via per-agent config entries.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
Use my subscription: Yes cross-checked
Use my subscription
Typed value: Yes.
External agents run on the user’s own subscriptions or API keys; provider keys also drive timeline Q&A and memory extraction; paid plans add included AI ($8/$40/$80) via granted AI access; local embedding model only; no multi-account switching documented.
- Agent subscriptions: Claude Code, Codex and other ACP agents run on the user’s own subscription or keys, including in a local org; CLI agents keep their own sign-in and subscription.
- Vendor credits: Paid plans include AI each month ($8/$40/$80) under granted AI access with monthly/daily budgets; no standalone credit purchase documented.
- Provider restrictions: uvx-only registry entries absent; binaries need a platform build; gateway refuses org-disabled models; per-agent first-run approvals (Codex trust, Cursor MCP approval, Pi trust).
- No local-LLM agent route or multi-account switching documented.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
Where agents run: On your computer: Yes cross-checked
Where agents run: On your computer
Typed value: Yes.
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
- Local: Every agent session runs on a team member’s own computer (desktop or CLI).
- Managed cloud: Ask Atlas Q&A/research agents run on Atlas servers; coding sessions have no cloud runtime (Issues/cloud agents roadmapped).
- No vendor-cloud coding runtime; cross-device use needs the desktop running.
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
Where agents run: On your own server: Yes cross-checked
Where agents run: On your own server
Typed value: Yes.
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
- Local: Every agent session runs on a team member’s own computer (desktop or CLI).
- Byo remote: CLI on a server reached over SSH records/syncs the same way (--no-browser sign-in).
- No vendor-cloud coding runtime; cross-device use needs the desktop running.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
Where agents run: In the vendor's cloud: Yes cross-checked
Where agents run: In the vendor's cloud
Typed value: Yes.
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
- Managed cloud: Ask Atlas Q&A/research agents run on Atlas servers; coding sessions have no cloud runtime (Issues/cloud agents roadmapped).
- Hybrid: Local execution with cloud-synced record plus server-side Ask Atlas; source never syncs.
- No vendor-cloud coding runtime; cross-device use needs the desktop running.
-
Finding: Organization's own agent on Atlas servers (no repo access): answers from recorded sessions/projects/comments/chat with links; drafts actions for approval; runs Deep research, Facts, Loops. Admin-enabled per org; billed to org AI budget with per-person daily question counts (e.g. "12 of 200 today"). Private per-person sessions; What-it-remembers deletable.
Workspace isolation: Git worktree: Yes
Workspace isolation: Git worktree
Typed value: Yes.
The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
- Branch: Per-session branches; branch switcher with uncommitted-change and in-use guards; update/rebase offers.
- Worktree: Fix/checkout flows use separate worktrees; per-worktree session starts; Fix worktrees auto-removed when settled/old/clean.
- Secrets: Secrets scrubbed on-machine before storage; sensitive env kept in keychain; Atlas never asks for its own gh token.
- Code separation: Worktree isolation for fixes/checkouts; concurrent sessions otherwise share the checkout with guards.
- Sandbox/container/VM internals undisclosed; no security-containment proof.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
Workspace isolation: Container: No
Workspace isolation: Container
Typed value: No.
The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
- Container: Built-in engine dispatches native local OS sandbox helpers, without a container runtime backend.
- Vm: No VM/microVM backend in the built-in engine launch configuration.
- Sandbox/container/VM internals undisclosed; no security-containment proof.
-
Atlas crates/atlas-native-agent/src/engine/config.rs
Finding: The built-in Atlas Agent launches local engine processes under OS policies: macOS Seatbelt, Linux sandbox helper and Windows restricted execution. The complete engine/deployment configuration contains no product container or VM worker runtime; separate Git worktrees handle file separation. External ACP-agent implementation choices are separate.
Workspace isolation: VM or microVM: No
Workspace isolation: VM or microVM
Typed value: No.
The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
- Container: Built-in engine dispatches native local OS sandbox helpers, without a container runtime backend.
- Vm: No VM/microVM backend in the built-in engine launch configuration.
- Sandbox/container/VM internals undisclosed; no security-containment proof.
-
Atlas crates/atlas-native-agent/src/engine/config.rs
Finding: The built-in Atlas Agent launches local engine processes under OS policies: macOS Seatbelt, Linux sandbox helper and Windows restricted execution. The complete engine/deployment configuration contains no product container or VM worker runtime; separate Git worktrees handle file separation. External ACP-agent implementation choices are separate.
Parallel runs: Yes cross-checked
Parallel runs
Typed value: Yes.
Multiple sessions run concurrently (second session beside the first; threads at once; watched PRs keep sessions working in background) with per-thread concurrent-run quotas on paid plans; sidebar and timeline show running/waiting/settled/live states. No observed concurrency fixture was run.
- Concurrent sessions: Second session beside the first; threads at once claimed with independent session state.
- Status visibility: Sidebar project/branch/status rows, live/needs-attention/finished marks, board auto-refresh.
- Concurrency is vendor-described; no observed fixture run.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
Coordinator: Yes
Coordinator
Typed value: Yes.
The built-in Atlas Agent supplies native child-agent delegation and communication. Atlas explicitly enables its multi_agent_v2 engine tools, supplies child instructions and exposes child activity; ACP third-party-agent behavior remains agent-specific.
- Roles: Native parent agent can spawn children for delegated work, with a cap of three threads per session.
- Messages: Native engine ships send_message and followup_task tools for child sessions.
- Orchestration owner: Atlas native agent config enables the product engine multi_agent_v2 tools and supplies developer instructions for delegated children.
- Live progress: list_agents/wait tools and SubAgentActivity events report child state.
- No inspectable parent/child delegation or cross-harness orchestration.
-
Atlas crates/atlas-native-agent/src/engine/config.rs
Finding: Atlas own native agent explicitly enables multi_agent_v2, supplies child developer instructions, and caps parent-plus-child concurrency at three. The shipped engine tools implement spawning, listing, messaging, follow-up, interruption and waiting on child agents. This is the product native Atlas Agent, not a displayed third-party CLI subagent feature.
Native issue tracking: No cross-checked
Native issue tracking
Typed value: No.
Native issue management does not exist in Atlas yet: Issues are roadmapped (dispatch agents to cloud machines, team-approved plans, sessions/commits attached). The timeline/session list is a session record, not an issue system; no parking, dispatch, states, or metadata exist.
- Create: Create: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Edit: Edit: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Discuss: Discuss: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Prioritize: Prioritize: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Label: Label: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Assign: Assign: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Search: Search: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Filter: Filter: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Archive: Archive: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Reopen: Reopen: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Project epic views: Project epic views: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Idea parking: Idea parking: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Explicit dispatch: Explicit dispatch: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Task session separation: Task session separation: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Metadata: Metadata: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- Durable states: Durable states: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
- External tracker requirement: External tracker requirement: not in Atlas yet; native issues are roadmapped with no shipped create/edit/discuss/state/metadata capability.
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
Unattended triggers: Yes cross-checked
Unattended triggers
Typed value: Yes.
No schedules or recurring tasks. Watched pull requests poll every 2 minutes and event-message the agent on check fail/pass, comments, reviews, and conflicts; incoming chat webhooks post external events to channels. Desktop must stay running (window may hide).
- Events: PR check/comment/review/conflict events message the watched session; incoming webhooks post CI/deploy/monitor events to channels.
- Completion conditions: Watches end on merge/close/settle/stop/8 read failures; idle sessions auto-settle after 3 days.
- Scheduler location: Watch polling runs in the desktop app, working while its window is hidden.
- Retries: 8 consecutive read failures end a watch (rate limits only delay); capture retry/requeue flows.
- Client independence: Works with window hidden; quitting the app ends runs (quit guard asks first).
- No schedules or recurring tasks; automation is event polling tied to the running desktop.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
Shared prompting: Yes cross-checked
Shared prompting
Typed value: Yes.
Shared timeline, threaded comments, live presence, co-written drafts/Spaces, and single-driver remote rooms with takeover; shared threads promise simultaneous co-work (in progress). No two-account co-steering test was run.
- Presence: Live session marks, room People panel (driving/here now), online members on dashboard.
- Coprompting: One driver at a time with handover; shared threads target simultaneous co-work but desktop sharing is in progress.
- Ordering: Driver seat serializes prompts; /queue and turn-end ordering for queued/watch messages.
- Permissions: Room invites, Let-teammates-drive toggle, private channels/DMs excluded even for admins.
- Owner online: Rooms execute on the sharer’s running desktop; Ask Atlas and cloud records need no owner online.
- Single-driver steering only; simultaneous co-work in progress; no two-account test.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
Team roles: Yes cross-checked
Team roles
Typed value: Yes.
Four org roles with a permission table, invite links, member removal, per-device sign-out, and plan-scoped surfaces (ROI, meetings). No enterprise SSO, SCIM, or unauthorized-access testing evidenced.
- Organization roles: Admin / Product Owner / Developer / Member with documented capabilities; Owner overlay (create→owner, no transfer).
- Membership: Email/invite-link invitations with Invitations tab; remove members; join via web accept.
- Revocation: Remove members; per-device remote sign-out; owner session deletion; stop sharing links.
- No SSO/SCIM; revocation is membership/device/session level; no access testing.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
Pull request workflow: Yes cross-checked
Pull request workflow
Typed value: Yes.
Full GitHub PR workflow inside Atlas: branch/worktree checkout, Create PR, review threads, checks, rebase/merge conflict resolution, revert, branch delete — all over gh with Atlas never pushing for the user.
- Branches: Branch switcher/create/delete/merge; per-session branches; PR stacks with chain view.
- Worktrees: Checkout in separate worktree; per-worktree session starts; Fix worktrees auto-cleaned.
- Pull requests: Create PR from session; tab opens on GitHub confirmation; fork-branch linking rules.
- Comments: Line comments, threads, held reviews, reactions, edit/delete where GitHub permits.
- Ci response: Checks listed fail-first with logs, stale verdicts, required-check pass events to agents.
- Rebase: Update-branch merge/rebase offers; checkpoint re-link across rebases.
- Conflicts: Resolve-conflicts opens an agent session; ours/theirs file resolution.
- Product state: Integrated PR tabs beside sessions with badges, tab persistence, and background refresh — not bare git commands.
- Push stays manual by design; authority ultimately GitHub-side.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
Review interface: Yes cross-checked
Review interface
Typed value: Yes.
Diffs in session review panels and PR Code view; call recordings/transcripts; session/memory/usage exports; timeline search retrieves post-session work with commit links. No dedicated artifact gallery or prompt-leak assessment.
- Diffs: Session diff review panel per project; PR Code view with reading order, per-commit picker, Viewed marks synced to GitHub.
- Documents: Guides, KB notes, canvas pages, call notes; no unified document-artifact store.
- Reviewer outputs: Guide tours and review verdicts retained per PR/commit.
- Revision links: Checkpoints tie sessions to exact commits; Timeline commit links; guides pinned per commit.
- Persistence: Settled/imported sessions reopenable; guides retained; share links open full record.
- No artifact gallery; prompt-leak behavior unassessed.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
Product license: Apache-2.0: Yes. Apache-2.0 cross-checked
Product license: Apache-2.0
Typed value: Yes. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: MIT: No. Apache-2.0 cross-checked
Product license: MIT
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: GPL-3.0: No. Apache-2.0 cross-checked
Product license: GPL-3.0
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: GPL-3.0-or-later: No. Apache-2.0 cross-checked
Product license: GPL-3.0-or-later
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: AGPL-3.0: No. Apache-2.0 cross-checked
Product license: AGPL-3.0
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: AGPL-3.0-or-later: No. Apache-2.0 cross-checked
Product license: AGPL-3.0-or-later
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: Elastic License 2.0: No. Apache-2.0 cross-checked
Product license: Elastic License 2.0
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: Custom terms: No. Apache-2.0 cross-checked
Product license: Custom terms
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Product license: Proprietary: No. Apache-2.0 cross-checked
Product license: Proprietary
Typed value: No. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Self-hosting: Yes cross-checked
Self-hosting
Typed value: Yes.
Local-first: desktop/CLI run on user machines with source builds documented; session/memory/usage exports in MD/JSON/PDF; org leave/delete and CLI uninstall remove vendor state. No self-hosted control plane or whole-org export.
- Self hosting: Desktop and CLI run on user machines; contributing docs cover source builds; cloud org stays vendor-hosted.
- Cancellation: Leave/delete organizations, per-device sign-out, CLI uninstall (keeps .atlas + sign-in for desktop).
- Account independence: Local org and CLI --local work without account/sync; team/cloud features need the vendor.
- No self-hosted org control plane; no whole-org portability.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
Telemetry opt-out: Yes cross-checked
Telemetry opt-out
Typed value: Yes.
On-machine redaction before storage, US hosting, owner-controlled deletion, and a complete pinned telemetry catalog with opt-out toggles; source builds inert without a key. Training use, subprocessors beyond PostHog/models.dev, and full retention policy undisclosed.
- Subprocessors: PostHog (telemetry) and models.dev (prices) named; no subprocessor list published.
- Telemetry disclosure: Pinned TELEMETRY.md catalogs every event, property and never-collected list; settings Privacy page describes sharing.
- Telemetry defaults: Share-usage defaults ON (opt-out, immediate); account-linkage toggle separate; in-app feedback sends even when off.
- Telemetry controls: Usage toggle + linkage toggle; keyless source builds are inert; per-kind notification switches separate.
- Training use, full subprocessor list, retention schedule, and first-run prompt undisclosed.
-
Finding: Complete telemetry catalog: emitter src-tauri/src/telemetry/mod.rs, backend PostHog; toggle Settings → General → "Share usage data" defaults ON (opt-out, immediate); second toggle for account linkage; signed-in usage attributed to account (deliberate change in 0.2.4); source builds without a PostHog key are inert; never collected: prompts, code, file paths, keys (per file).
Free access: Yes cross-checked
Free access
Typed value: Yes.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: Free: Yes. Free cross-checked
Starting price: Free
Typed value: Yes. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: Up to $20/month: No. Free cross-checked
Starting price: Up to $20/month
Typed value: No. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: $21–40/month: No. Free cross-checked
Starting price: $21–40/month
Typed value: No. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: $41–100/month: No. Free cross-checked
Starting price: $41–100/month
Typed value: No. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: Over $100/month: No. Free cross-checked
Starting price: Over $100/month
Typed value: No. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Starting price: Contact for pricing: No. Free cross-checked
Starting price: Contact for pricing
Typed value: No. Free.
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Ongoing free: Free $0/month plan: 3 members, unlimited synced projects per table, Standard storage/meetings, shared threads.
- Paid only: A standing Free plan and free download exist, so access is not paid-only.
- Free scenarios: Two developers fit Free seats with chat/timeline/sessions; metered AI and meetings need paid tiers or own subs.
- Usage caps: Member, storage, vectorization, run-concurrency and AI-allowance caps published per tier.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Free, no paid plan: No cross-checked
Pricing model: Free, no paid plan
Typed value: No.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Free plan: Yes cross-checked
Pricing model: Free plan
Typed value: Yes.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Subscription: Yes cross-checked
Pricing model: Subscription
Typed value: Yes.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Per seat: No
Pricing model: Per seat
Typed value: No.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Usage-based: No
Pricing model: Usage-based
Typed value: No.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Per machine or compute: No
Pricing model: Per machine or compute
Typed value: No.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Pricing model: Bring your own key: Yes cross-checked
Pricing model: Bring your own key
Typed value: Yes.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
Pricing model: One-time purchase: No
Pricing model: One-time purchase
Typed value: No.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: No paid plan: No. $11–20/month
Paid plans from: No paid plan
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: $10 or less/month: No. $11–20/month
Paid plans from: $10 or less/month
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: $11–20/month: Yes. $11–20/month
Paid plans from: $11–20/month
Typed value: Yes. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: $21–40/month: No. $11–20/month
Paid plans from: $21–40/month
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: $41–100/month: No. $11–20/month
Paid plans from: $41–100/month
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: Over $100/month: No. $11–20/month
Paid plans from: Over $100/month
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plans from: Contact sales: No. $11–20/month
Paid plans from: Contact sales
Typed value: No. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Keeps going with laptop closed: No cross-checked
Keeps going with laptop closed
Typed value: No.
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
- Client disconnect: Rooms survive tab close/project switch and watches run with window hidden; quitting the desktop ends runs.
- Other device continuation: Web app and phone-browser rooms view/steer the same live session; execution stays on the desktop.
- No vendor-cloud coding runtime; cross-device use needs the desktop running.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
Open-source license: Yes. Apache-2.0 cross-checked
Open-source license
Typed value: Yes. Apache-2.0.
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- Client license: Root LICENSE is Apache License 2.0; API reports Apache-2.0; README badges it.
- CLI repo license and hosted-service terms unverified.
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
Any desktop app: Yes cross-checked
Any desktop app
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Desktop os: macOS 11+ Apple silicon/Intel, Windows 10+ x64, Linux x86-64 packages (deb/rpm/Arch/AppImage/tarball; WebKitGTK 4.1).
- Cli: atlas-cli on macOS and Linux for capture/sync/memory (not on Windows); desktop also installs an `atlas` folder-opener on macOS and `atl`/`tryatlas` on Linux.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
Paid plan up to $20/month: Yes. $11–20/month
Paid plan up to $20/month
Typed value: Yes. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Paid plan up to $40/month: Yes. $11–20/month
Paid plan up to $40/month
Typed value: Yes. $11–20/month.
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Currency: US dollars on the public pricing table.
- Interval: Per-month prices ($0/$20/$100/$200 per month).
- Seats: Member allowances 3/5/25/100 per tier ("Plans add seats"); per-seat vs pooled billing unstated.
- Included usage: Included monthly AI ($0/$8/$40/$80), vectorization tokens, session storage, concurrent runs, meetings.
- Trial: Free plan plus free-during-alpha framing and beta program; no time-boxed trial terms stated.
- Model cost: BYOK supported with own keys; included AI allowance; no pass-through unit rates shown.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
Any phone client: Yes cross-checked
Any phone client
Typed value: Yes.
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- Browser: Released web app at app.tryatlas.cc: inbox, timeline, projects, chat, ROI, settings, share links and live remote-control rooms with prompt/stop/model controls. Runs no coding agents.
- Mobile browser: Remote-control room page is built for a phone screen and carries the live session with prompt/stop/model controls; no native app needed for it.
- Task actions: Desktop starts/steers sessions (second session beside it, worktree starts, settle/snooze), approvals via permission prompts, review in diffs/PR tabs, resume repaints transcript from disk; web rooms prompt/stop/switch model.
- Cross device continuity: Remote-control rooms and org sync continue the same session from another computer or phone browser, but the agent keeps running on the desktop; "pick up work on any device" is roadmapped.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
All public findings
Findings may summarize a source. They are not presented as verbatim statements.
atlas:ACC-01: Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop. cross-checked
atlas:ACC-01
Desktop app on macOS 11+ (ARM/Intel), Windows 10+ x64 and Linux x86-64; released web app (browser, incl. a phone-built remote-control room page) with prompt/stop/model controls; atlas-cli on macOS/Linux (no Windows CLI); no native phone app yet. Desktop sessions start/steer/approve/review/resume; notifications via native banners, in-app center and email; cross-device handoff carries the live session but execution stays on the desktop.
- No native mobile app; CLI absent on Windows; cross-device use needs the originating desktop running.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: No dates given ("we don't commit to them"). Shipped: ACP agents, session capture/checkpoints, team timeline 0.3.4, Atlas Agent in org 0.3.4, team chat 0.3.1, calls 0.3.1, prompt drafts 0.3.1, Spaces 0.3.1, shared threads, AI gateway/budgets, usage dashboard, AI ROI, remote control 0.4.1, Ask Atlas 0.4.1, shared memory, Atlas CLI, Linux 0.3.4. In progress: Atlas 0.4.0 (notifications, keymaps, agent switching, /remember), mobile app (iOS/Android client, no agent execution). Roadmapped: Issues and cloud agents, Self-healing, Handbook, "Pick up work on any device".
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: "In progress: being built, and not in a release yet" and "not in the App Store or Google Play yet." Planned iOS/Android client for chat, timeline, inbox, projects, members, dashboard; no agent execution; no calls (join from web/desktop); no push notifications; no Spaces/drafts/shared threads. Device-code sign-in like desktop.
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
atlas:ACC-02: Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use. cross-checked
atlas:ACC-02
Atlas Agent built in plus ACP-registry agents (Claude Code, Codex, Cursor, OpenCode and more) in one composer with per-session agent/model/mode choice; atlas-cli hooks 8 terminal agents incl. Copilot CLI, Factory Droid, Antigravity and Pi. BYOK provider keys (Cheaper Inference, OrcaRouter, Empero named) and custom model names; sessions controlled through ACP/hooks with side-by-side mixed-agent use.
- Mixed-harness operation is vendor-described; no observed two-harness fixture was run.
-
Atlas — Parallel agents section
Finding: Vendor claims a team of agents can run at once with true concurrency: "Rust-owned session state means threads run independently", mixing Claude Code, Codex and ACP agents side by side, intervening in any running thread.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Official ACP catalog in Settings → Agents; runs registry entries publishing an npx package or a platform binary (uvx-only entries absent; binary-only entries visible but not installable without a build for the OS). Atlas manages its own Node runtime, per-agent install dirs; Update button on newer versions; Remove deletes managed cache, sessions stay. Transport-integrity trust signal (checksums), not a security review.
-
Finding: 0.4.1 unreleased/in development (remote control, Ask Atlas, appearance). 0.4.0 unreleased (notifications pipeline with Allow once/Deny banners, keymap presets, agent switching keeps conversation, /remember, AGENTS.md mirroring, Cheaper Inference BYOK). 0.3.4 2026-09-27 Teams Early (cloud timeline, comments, org Atlas Agent, Linux, OrcaRouter BYOK). 0.3.3 2026-09-19, 0.3.2 2026-09-16 (first Windows build), 0.3.1 2026-09-06 (comms alpha, Atlas Agent billing, slash commands, config.toml), 0.3.0 2026-08-25 (ACP + timeline). Full history on GitHub.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
atlas:ACC-03: External agents run on the user’s own subscriptions or API keys; provider keys also drive timeline Q&A and memory extraction; paid plans add included AI ($8/$40/$80) via granted AI access; local embedding model only; no multi-account switching documented. cross-checked
atlas:ACC-03
External agents run on the user’s own subscriptions or API keys; provider keys also drive timeline Q&A and memory extraction; paid plans add included AI ($8/$40/$80) via granted AI access; local embedding model only; no multi-account switching documented.
- No local-LLM agent route or multi-account switching documented.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Official ACP catalog in Settings → Agents; runs registry entries publishing an npx package or a platform binary (uvx-only entries absent; binary-only entries visible but not installable without a build for the OS). Atlas manages its own Node runtime, per-agent install dirs; Update button on newer versions; Remove deletes managed cache, sessions stay. Transport-integrity trust signal (checksums), not a security review.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
atlas:ACC-04: Install to first team-visible session is guided (install, add project from 4 sources, add agent, capture on; Get-started checklist). Folders need no git; GitHub sources need gh. Managed-cloud execution does not exist (roadmapped cloud agents); a CLI-on-server path over SSH is documented. cross-checked
atlas:ACC-04
Install to first team-visible session is guided (install, add project from 4 sources, add agent, capture on; Get-started checklist). Folders need no git; GitHub sources need gh. Managed-cloud execution does not exist (roadmapped cloud agents); a CLI-on-server path over SSH is documented.
- Managed cloud absent; VPS path documented but unmeasured.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: Fresh install to first session: install Atlas, add a project (New/Local folder/Git URL/GitHub owner-repo via gh), add an agent (Atlas Agent needs synced org with AI access; otherwise install from ACP registry), turn on capture; second half connects the team via synced organization.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
atlas:ACC-05: atlas-cli covers account/org/setup/agents/capture/sync/memory/status/update; desktop installs a folder-opener; incoming chat webhooks and a memory MCP server/bridge exist; no documented public REST API operations, SDK, or outgoing webhooks; personal access token for scripts exists without documented operations. cross-checked
atlas:ACC-05
atlas-cli covers account/org/setup/agents/capture/sync/memory/status/update; desktop installs a folder-opener; incoming chat webhooks and a memory MCP server/bridge exist; no documented public REST API operations, SDK, or outgoing webhooks; personal access token for scripts exists without documented operations.
- No public REST API operations, SDK, or outgoing webhooks documented.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
atlas:ACC-06: Skills and skills.sh packs linked into Atlas/Claude Code/Codex targets; instruction files mirrored (AGENTS.md/CLAUDE.md); notes importable and @-mentionable; memory MCP server for capable agents; registry checksum trust signal; per-target skill enablement and sandbox bounds. No external MCP-server connections, plugins, or custom tool authoring documented. cross-checked
atlas:ACC-06
Skills and skills.sh packs linked into Atlas/Claude Code/Codex targets; instruction files mirrored (AGENTS.md/CLAUDE.md); notes importable and @-mentionable; memory MCP server for capable agents; registry checksum trust signal; per-target skill enablement and sandbox bounds. No external MCP-server connections, plugins, or custom tool authoring documented.
- No MCP-client connections, plugin installs, or custom tool authoring documented.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Official ACP catalog in Settings → Agents; runs registry entries publishing an npx package or a platform binary (uvx-only entries absent; binary-only entries visible but not installable without a build for the OS). Atlas manages its own Node runtime, per-agent install dirs; Update button on newer versions; Remove deletes managed cache, sessions stay. Transport-integrity trust signal (checksums), not a security review.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Skills + packs (skills.sh registry) managed in Settings → Skills; canonical store ~/.agents/skills (global) and <project>/.agents/skills; linked into enabled targets Atlas/Claude Code/Codex; external agent-dir skills listed read-only. Invoke via / in composer using what the agent advertises. Atlas ships remember and atlas-self-configure skills.
-
Finding: Policy tab reads CLAUDE.md, AGENTS.md and Claude memory files into a table of 12 preference kinds with source text, file and match strength; edits rewrite the same span in the original file; Strong/Soft classification; Send a rule to agent chat.
-
Finding: Markdown notes with blocks, backlinks, @ file mentions, icons/covers/properties; graph view; import folders (e.g. Obsidian vaults). Attach via @ in composer as visible context block; memory-searchable by agents.
atlas:ACC-07: Desktop is Tauri (pinned tauri.conf.json; per-OS configs; "no Electron" vendor claim); mobile client unreleased with undisclosed implementation. cross-checked
atlas:ACC-07
Desktop is Tauri (pinned tauri.conf.json; per-OS configs; "no Electron" vendor claim); mobile client unreleased with undisclosed implementation.
- Mobile implementation undisclosed; no hybrid embedding documented.
-
Finding: v0.4.1; macOS 13+ (Apple Silicon and Intel), Windows 10+ (x64); "Native binary for Apple Silicon, Intel and Windows, no Electron"; "The Atlas account is optional"; address Atlas 1111B S Governors Ave #39146 Dover, DE 19904; links to GitHub, X (@tryatlas_cc) and Discord.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: "In progress: being built, and not in a release yet" and "not in the App Store or Google Play yet." Planned iOS/Android client for chat, timeline, inbox, projects, members, dashboard; no agent execution; no calls (join from web/desktop); no push notifications; no Spaces/drafts/shared threads. Device-code sign-in like desktop.
-
src-tauri/tauri.conf.json at pinned commit
Finding: Tauri app config: productName "Atlas", identifier "dev.atlas.ide", version 0.4.1, desktop window 1280x800 with overlay title bar; per-OS tauri.*.conf.json (macos/windows/linux) present.
atlas:PLAN-01: Sessions and projects are created locally with prompt intake; terminal sessions importable; checkpoint-stable IDs with cloud sync; watched-PR and chat-webhook events feed agents. No external tracker import or authoritative-tracker story documented. cross-checked
atlas:PLAN-01
Sessions and projects are created locally with prompt intake; terminal sessions importable; checkpoint-stable IDs with cloud sync; watched-PR and chat-webhook events feed agents. No external tracker import or authoritative-tracker story documented.
- No external tracker import/sync story; event triggers message agents rather than dispatch tracked work.
-
Finding: Fresh install to first session: install Atlas, add a project (New/Local folder/Git URL/GitHub owner-repo via gh), add an agent (Atlas Agent needs synced org with AI access; otherwise install from ACP registry), turn on capture; second half connects the team via synced organization.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: Live co-edited prompt documents per conversation (named cursors, avatars; voice call from draft in web app); Send to agent runs on one person's machine with their agent/credentials/spend; nothing runs on the server.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Multiplayer defined as shared record: timeline of every recorded session with prompts/tool calls/commits, comments beside the work, team chat/calls/drafts/Spaces/shared threads; organization owns sharing; cloud projects merge teammates' sessions; source code never syncs.
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:PLAN-02: Sessions persist as durable transcripts with memory Plan entries and commit checkpoints; reopening repaints history and messaging a settled session returns it. No acceptance-criteria construct, plan editor, or revision history documented. cross-checked
atlas:PLAN-02
Sessions persist as durable transcripts with memory Plan entries and commit checkpoints; reopening repaints history and messaging a settled session returns it. No acceptance-criteria construct, plan editor, or revision history documented.
- Plans are memory entries + transcripts; no acceptance criteria, editors, or revisions.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:PLAN-03: Sessions keep identity across restarts, settle-on-merge propagates PR state, checkpoints track history rewrites, and PR stacks render parent/child chains. No epic/hierarchy, blocker, readiness, or enforcement model documented. cross-checked
atlas:PLAN-03
Sessions keep identity across restarts, settle-on-merge propagates PR state, checkpoints track history rewrites, and PR stacks render parent/child chains. No epic/hierarchy, blocker, readiness, or enforcement model documented.
- No epics, blockers, readiness, or dependency enforcement; state follows sessions/PRs only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:PLAN-04: Per-repo shared memory (decisions/facts/failures/architecture/plans) with provenance, edit/forget, verify and cross-project promotion; Policy table over instruction files; knowledge-base notes; code index with local embeddings. Teammate sync of memory absent (local .atlas only); stale controls partial. cross-checked
atlas:PLAN-04
Per-repo shared memory (decisions/facts/failures/architecture/plans) with provenance, edit/forget, verify and cross-project promotion; Policy table over instruction files; knowledge-base notes; code index with local embeddings. Teammate sync of memory absent (local .atlas only); stale controls partial.
- Memory is per-machine (not team-synced); no staleness expiry documented.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Skills + packs (skills.sh registry) managed in Settings → Skills; canonical store ~/.agents/skills (global) and <project>/.agents/skills; linked into enabled targets Atlas/Claude Code/Codex; external agent-dir skills listed read-only. Invoke via / in composer using what the agent advertises. Atlas ships remember and atlas-self-configure skills.
-
Finding: Policy tab reads CLAUDE.md, AGENTS.md and Claude memory files into a table of 12 preference kinds with source text, file and match strength; edits rewrite the same span in the original file; Strong/Soft classification; Send a rule to agent chat.
-
Finding: Markdown notes with blocks, backlinks, @ file mentions, icons/covers/properties; graph view; import folders (e.g. Obsidian vaults). Attach via @ in composer as visible context block; memory-searchable by agents.
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
atlas:PLAN-05: Project code index with local embeddings grounds memory search; Ask Atlas answers with links; comments anchor to exact lines; chat accepts images but Atlas Agent is text-only. No hands-on retrieval/citation/conflict test was run. cross-checked
atlas:PLAN-05
Project code index with local embeddings grounds memory search; Ask Atlas answers with links; comments anchor to exact lines; chat accepts images but Atlas Agent is text-only. No hands-on retrieval/citation/conflict test was run.
- No hands-on retrieval/citation/conflict scenario was run; external knowledge sourcing unclear.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Finding: Organization's own agent on Atlas servers (no repo access): answers from recorded sessions/projects/comments/chat with links; drafts actions for approval; runs Deep research, Facts, Loops. Admin-enabled per org; billed to org AI budget with per-person daily question counts (e.g. "12 of 200 today"). Private per-person sessions; What-it-remembers deletable.
atlas:PLAN-06: Native issue management does not exist in Atlas yet: Issues are roadmapped (dispatch agents to cloud machines, team-approved plans, sessions/commits attached). The timeline/session list is a session record, not an issue system; no parking, dispatch, states, or metadata exist. cross-checked
atlas:PLAN-06
Native issue management does not exist in Atlas yet: Issues are roadmapped (dispatch agents to cloud machines, team-approved plans, sessions/commits attached). The timeline/session list is a session record, not an issue system; no parking, dispatch, states, or metadata exist.
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
-
Finding: No dates given ("we don't commit to them"). Shipped: ACP agents, session capture/checkpoints, team timeline 0.3.4, Atlas Agent in org 0.3.4, team chat 0.3.1, calls 0.3.1, prompt drafts 0.3.1, Spaces 0.3.1, shared threads, AI gateway/budgets, usage dashboard, AI ROI, remote control 0.4.1, Ask Atlas 0.4.1, shared memory, Atlas CLI, Linux 0.3.4. In progress: Atlas 0.4.0 (notifications, keymaps, agent switching, /remember), mobile app (iOS/Android client, no agent execution). Roadmapped: Issues and cloud agents, Self-healing, Handbook, "Pick up work on any device".
atlas:PLAN-07: No durable issue IDs exist for agents to operate (Issues roadmapped). Agents do operate adjacent records via tools (watch/link/list pull requests, PR-watch messages), but shared authoritative issue records with human-visible writes are absent. cross-checked
atlas:PLAN-07
No durable issue IDs exist for agents to operate (Issues roadmapped). Agents do operate adjacent records via tools (watch/link/list pull requests, PR-watch messages), but shared authoritative issue records with human-visible writes are absent.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
atlas:PLAN-08: Knowledge-base notes and memory decisions persist with provenance and retrieval, but the team-agreed versioned spec (Handbook) is roadmapped; no approval workflow, requirement links, or drift detection exist. cross-checked
atlas:PLAN-08
Knowledge-base notes and memory decisions persist with provenance and retrieval, but the team-agreed versioned spec (Handbook) is roadmapped; no approval workflow, requirement links, or drift detection exist.
- No versioned team spec, approvals, requirement links, or drift detection; Handbook is future.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Policy tab reads CLAUDE.md, AGENTS.md and Claude memory files into a table of 12 preference kinds with source text, file and match strength; edits rewrite the same span in the original file; Strong/Soft classification; Send a rule to agent chat.
-
Finding: Markdown notes with blocks, backlinks, @ file mentions, icons/covers/properties; graph view; import folders (e.g. Obsidian vaults). Attach via @ in composer as visible context block; memory-searchable by agents.
atlas:COORD-01: Multiple sessions run concurrently (second session beside the first; threads at once; watched PRs keep sessions working in background) with per-thread concurrent-run quotas on paid plans; sidebar and timeline show running/waiting/settled/live states. No observed concurrency fixture was run. cross-checked
atlas:COORD-01
Multiple sessions run concurrently (second session beside the first; threads at once; watched PRs keep sessions working in background) with per-thread concurrent-run quotas on paid plans; sidebar and timeline show running/waiting/settled/live states. No observed concurrency fixture was run.
- Concurrency is vendor-described; no observed fixture run.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Atlas — Parallel agents section
Finding: Vendor claims a team of agents can run at once with true concurrency: "Rust-owned session state means threads run independently", mixing Claude Code, Codex and ACP agents side by side, intervening in any running thread.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:COORD-02: Sessions run under a chosen agent with per-project defaults; PR/message queues order work; the driver seat and checkout locks serialize access. No claim/duplicate-prevention or routing model documented. cross-checked
atlas:COORD-02
Sessions run under a chosen agent with per-project defaults; PR/message queues order work; the driver seat and checkout locks serialize access. No claim/duplicate-prevention or routing model documented.
- No claims, routing, or duplicate prevention; ownership is seat/checkout serialization only.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Live co-edited prompt documents per conversation (named cursors, avatars; voice call from draft in web app); Send to agent runs on one person's machine with their agent/credentials/spend; nothing runs on the server.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:COORD-03: The built-in Atlas Agent supplies native child-agent delegation and communication. Atlas explicitly enables its multi_agent_v2 engine tools, supplies child instructions and exposes child activity; ACP third-party-agent behavior remains agent-specific.
atlas:COORD-03
The built-in Atlas Agent supplies native child-agent delegation and communication. Atlas explicitly enables its multi_agent_v2 engine tools, supplies child instructions and exposes child activity; ACP third-party-agent behavior remains agent-specific.
- No inspectable parent/child delegation or cross-harness orchestration.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Hook table: Claude Code, Codex, Cursor CLI+editor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi with per-agent hook paths and memory-server paths; Atlas edits only its own entries and installs a git post-commit hook. First-run approvals per agent (e.g. Codex trust, Cursor approve atlas_memory server, Pi trust project). Every hooked agent records prompt, tool calls, files changed; stopped agents still sync.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Live co-edited prompt documents per conversation (named cursors, avatars; voice call from draft in web app); Send to agent runs on one person's machine with their agent/credentials/spend; nothing runs on the server.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
-
Atlas crates/atlas-native-agent/src/engine/config.rs
Finding: Atlas own native agent explicitly enables multi_agent_v2, supplies child developer instructions, and caps parent-plus-child concurrency at three. The shipped engine tools implement spawning, listing, messaging, follow-up, interruption and waiting on child agents. This is the product native Atlas Agent, not a displayed third-party CLI subagent feature.
atlas:COORD-04: No schedules or recurring tasks. Watched pull requests poll every 2 minutes and event-message the agent on check fail/pass, comments, reviews, and conflicts; incoming chat webhooks post external events to channels. Desktop must stay running (window may hide). cross-checked
atlas:COORD-04
No schedules or recurring tasks. Watched pull requests poll every 2 minutes and event-message the agent on check fail/pass, comments, reviews, and conflicts; incoming chat webhooks post external events to channels. Desktop must stay running (window may hide).
- No schedules or recurring tasks; automation is event polling tied to the running desktop.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
atlas:COORD-05: Running turns can be stopped, sessions steered mid-run with queued follow-ups, resumed from disk, and interrupted by approvals/questions; notifications surface finishes, mentions, and errors with reasons. No execution pause or escalation chain documented. cross-checked
atlas:COORD-05
Running turns can be stopped, sessions steered mid-run with queued follow-ups, resumed from disk, and interrupted by approvals/questions; notifications surface finishes, mentions, and errors with reasons. No execution pause or escalation chain documented.
- No execution pause or escalation chain; terminal CLI cannot steer/stop.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: 0.4.1 unreleased/in development (remote control, Ask Atlas, appearance). 0.4.0 unreleased (notifications pipeline with Allow once/Deny banners, keymap presets, agent switching keeps conversation, /remember, AGENTS.md mirroring, Cheaper Inference BYOK). 0.3.4 2026-09-27 Teams Early (cloud timeline, comments, org Atlas Agent, Linux, OrcaRouter BYOK). 0.3.3 2026-09-19, 0.3.2 2026-09-16 (first Windows build), 0.3.1 2026-09-06 (comms alpha, Atlas Agent billing, slash commands, config.toml), 0.3.0 2026-08-25 (ACP + timeline). Full history on GitHub.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
atlas:COORD-06: Timeline board groups the newest 500 sessions by day with live/attention marks, filters and search; PR lists group Authored/Review-requested/Others with filters/sorts and stack chains; usage/ROI dashboards show workload. No 12-state fixture was run, so navigation depth is undescribed. cross-checked
atlas:COORD-06
Timeline board groups the newest 500 sessions by day with live/attention marks, filters and search; PR lists group Authored/Review-requested/Others with filters/sorts and stack chains; usage/ROI dashboards show workload. No 12-state fixture was run, so navigation depth is undescribed.
- No fixture run; dependency and error-navigation behavior unmeasured.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: Web app ROI: spend vs commits/active time/lines changed by project/developer/model/session; Admin/Product Owner/owner only; covers readable projects; deleted sessions leave the figures.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:COORD-07: Quick sessions start instantly with per-agent/model/mode config and mid-run changes; PR actions show what ran with retained guides; discovery via palette/checklists/tips. No mandatory presets, NL workflow configuration, or epic growth path. cross-checked
atlas:COORD-07
Quick sessions start instantly with per-agent/model/mode config and mid-run changes; PR actions show what ran with retained guides; discovery via palette/checklists/tips. No mandatory presets, NL workflow configuration, or epic growth path.
- No presets, NL orchestration config, or epic growth path.
-
Finding: Fresh install to first session: install Atlas, add a project (New/Local folder/Git URL/GitHub owner-repo via gh), add an agent (Atlas Agent needs synced org with AI access; otherwise install from ACP registry), turn on capture; second half connects the team via synced organization.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: 0.4.1 unreleased/in development (remote control, Ask Atlas, appearance). 0.4.0 unreleased (notifications pipeline with Allow once/Deny banners, keymap presets, agent switching keeps conversation, /remember, AGENTS.md mirroring, Cheaper Inference BYOK). 0.3.4 2026-09-27 Teams Early (cloud timeline, comments, org Atlas Agent, Linux, OrcaRouter BYOK). 0.3.3 2026-09-19, 0.3.2 2026-09-16 (first Windows build), 0.3.1 2026-09-06 (comms alpha, Atlas Agent billing, slash commands, config.toml), 0.3.0 2026-08-25 (ACP + timeline). Full history on GitHub.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Live co-edited prompt documents per conversation (named cursors, avatars; voice call from draft in web app); Send to agent runs on one person's machine with their agent/credentials/spend; nothing runs on the server.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Default shortcuts incl. Cmd-K palette, Cmd-T surfaces, Cmd-Shift-B source control, Cmd-J terminal, Alt-/ agent cycle, Shift-Tab mode cycle; keymap presets VS Code/Cursor/Zed/JetBrains with import/export; layouts; hint navigation. Recent Sessions shows per-session project, local-vs-worktree, agent, recency.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:RUN-01: Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting. cross-checked
atlas:RUN-01
Coding sessions execute locally (desktop or CLI, incl. SSH servers); Ask Atlas runs on Atlas servers; no vendor-cloud coding runtime (cloud agents roadmapped). Remote rooms and the web app continue the same session cross-device while the desktop runs; US org hosting.
- No vendor-cloud coding runtime; cross-device use needs the desktop running.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: "Roadmapped: not in Atlas yet." Planned: issues that dispatch an agent to a cloud machine, team-approved plans, sessions and commits attached to the issue. Usable today: session capture, team timeline, comments, Atlas Agent. "Today every agent session in Atlas runs on a team member's own computer, and Atlas doesn't open pull requests."
-
Finding: No dates given ("we don't commit to them"). Shipped: ACP agents, session capture/checkpoints, team timeline 0.3.4, Atlas Agent in org 0.3.4, team chat 0.3.1, calls 0.3.1, prompt drafts 0.3.1, Spaces 0.3.1, shared threads, AI gateway/budgets, usage dashboard, AI ROI, remote control 0.4.1, Ask Atlas 0.4.1, shared memory, Atlas CLI, Linux 0.3.4. In progress: Atlas 0.4.0 (notifications, keymaps, agent switching, /remember), mobile app (iOS/Android client, no agent execution). Roadmapped: Issues and cloud agents, Self-healing, Handbook, "Pick up work on any device".
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Organization's own agent on Atlas servers (no repo access): answers from recorded sessions/projects/comments/chat with links; drafts actions for approval; runs Deep research, Facts, Loops. Admin-enabled per org; billed to org AI budget with per-person daily question counts (e.g. "12 of 200 today"). Private per-person sessions; What-it-remembers deletable.
atlas:RUN-02: The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
atlas:RUN-02
The built-in Atlas Agent uses worktrees and local operating-system sandbox policies. Its scoped engine launch paths contain no product-owned container or VM runtime; third-party ACP-agent internals remain outside this claim.
- Sandbox/container/VM internals undisclosed; no security-containment proof.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Atlas crates/atlas-native-agent/src/engine/config.rs
Finding: The built-in Atlas Agent launches local engine processes under OS policies: macOS Seatbelt, Linux sandbox helper and Windows restricted execution. The complete engine/deployment configuration contains no product container or VM worker runtime; separate Git worktrees handle file separation. External ACP-agent implementation choices are separate.
atlas:RUN-03: Transcripts reload from disk with agent reconnect; capture survives restarts with a repair table; watches persist across tab/project close. Crash, duplication, and multi-day-continuity behavior untested. cross-checked
atlas:RUN-03
Transcripts reload from disk with agent reconnect; capture survives restarts with a repair table; watches persist across tab/project close. Crash, duplication, and multi-day-continuity behavior untested.
- Crash/duplication/multi-day behavior untested; resume is transcript+reconnect only.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:RUN-04: Agents perform setup via commands; secrets via env/keychain/provider keys; CLI installer pulls deps with checksum; per-project memory/permission settings. Dependency install behavior, services, ports, and repair automation undocumented. cross-checked
atlas:RUN-04
Agents perform setup via commands; secrets via env/keychain/provider keys; CLI installer pulls deps with checksum; per-project memory/permission settings. Dependency install behavior, services, ports, and repair automation undocumented.
- Environment provisioning is agent-executed; no services/ports/snapshot management.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:RUN-05: Org monthly/daily AI budgets with per-member caps hard-stop new requests; context-full warnings; retry/requeue flows for capture, sync and GitHub rate limits; memory/provider call fallbacks. Time/turn/token hard limits undocumented. cross-checked
atlas:RUN-05
Org monthly/daily AI budgets with per-member caps hard-stop new requests; context-full warnings; retry/requeue flows for capture, sync and GitHub rate limits; memory/provider call fallbacks. Time/turn/token hard limits undocumented.
- Time/turn/token hard limits absent; failure handling described, not tested.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
-
Finding: Organization's own agent on Atlas servers (no repo access): answers from recorded sessions/projects/comments/chat with links; drafts actions for approval; runs Deep research, Facts, Loops. Admin-enabled per org; billed to org AI budget with per-person daily question counts (e.g. "12 of 200 today"). Private per-person sessions; What-it-remembers deletable.
atlas:RUN-06: Every session keeps prompts/tool calls/file changes with checkpoints; watched-PR and capture failures surface with reasons; usage/ROI break down tokens/cost by session/agent/project/model with exports. Reconnect consistency and performance unmeasured. cross-checked
atlas:RUN-06
Every session keeps prompts/tool calls/file changes with checkpoints; watched-PR and capture failures surface with reasons; usage/ROI break down tokens/cost by session/agent/project/model with exports. Reconnect consistency and performance unmeasured.
- No induced-failure or performance fixture; status agreement across reconnect untested.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Web app ROI: spend vs commits/active time/lines changed by project/developer/model/session; Admin/Product Owner/owner only; covers readable projects; deleted sessions leave the figures.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Own shell tabs beside sessions, splits, link finding, zsh integration; finish notifications (failures/input always; success past threshold; 10s default).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:RUN-07: Usage shows tokens (incl. cache/reasoning) and estimated cost per session, groupable by project/agent/model with report exports; org Atlas Agent spend tracked per person against budgets; provider-bill reconciliation, compute fees and budget alerts undocumented. cross-checked
atlas:RUN-07
Usage shows tokens (incl. cache/reasoning) and estimated cost per session, groupable by project/agent/model with report exports; org Atlas Agent spend tracked per person against budgets; provider-bill reconciliation, compute fees and budget alerts undocumented.
- Estimates only; no reconciliation, compute fees, refresh timing, or alerts.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Web app ROI: spend vs commits/active time/lines changed by project/developer/model/session; Admin/Product Owner/owner only; covers readable projects; deleted sessions leave the figures.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
atlas:RUN-08: Per-thread concurrent-run quotas bound shared-thread parallelism; message/watch queues serialize agent interruptions; cancellation (Stop/Stop-watch/uninstall) documented. CPU/RAM limits, backpressure, check dedup, and fairness undocumented; no load test run. cross-checked
atlas:RUN-08
Per-thread concurrent-run quotas bound shared-thread parallelism; message/watch queues serialize agent interruptions; cancellation (Stop/Stop-watch/uninstall) documented. CPU/RAM limits, backpressure, check dedup, and fairness undocumented; no load test run.
- Only quotas/queues/cancellation evidenced; no resource-aware scheduling proof.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
atlas:TEAM-01: Shared timeline, threaded comments, live presence, co-written drafts/Spaces, and single-driver remote rooms with takeover; shared threads promise simultaneous co-work (in progress). No two-account co-steering test was run. cross-checked
atlas:TEAM-01
Shared timeline, threaded comments, live presence, co-written drafts/Spaces, and single-driver remote rooms with takeover; shared threads promise simultaneous co-work (in progress). No two-account co-steering test was run.
- Single-driver steering only; simultaneous co-work in progress; no two-account test.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: Voice (audio, members, every plan) and video meetings (members + guest links, recording/transcript/AI note; plan-gated) started from any conversation, run in the web app; voice-only fallback when plan lacks meetings.
-
Finding: Live co-edited prompt documents per conversation (named cursors, avatars; voice call from draft in web app); Send to agent runs on one person's machine with their agent/credentials/spend; nothing runs on the server.
-
Finding: Shared realtime canvas per conversation (notes/text/frames/shapes/images/video, undo); Atlas Agent can draw on it; conversation-scoped. Distinct from local project canvas (.atlas/canvas.json, not shared).
-
Finding: "In progress: being built, and not in a release yet." Vision: shared checkout where teammates and agents type/run agents together with merged turns and resolved conflicts; thread view live in web app today (watch, resolve conflicts, compare/restore versions, comment, ask teammate's agent to run a prompt); sharing/typing/running/owner controls arriving in desktop.
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: Multiplayer defined as shared record: timeline of every recorded session with prompts/tool calls/commits, comments beside the work, team chat/calls/drafts/Spaces/shared threads; organization owns sharing; cloud projects merge teammates' sessions; source code never syncs.
-
Finding: Organization's own agent on Atlas servers (no repo access): answers from recorded sessions/projects/comments/chat with links; drafts actions for approval; runs Deep research, Facts, Loops. Admin-enabled per org; billed to org AI budget with per-person daily question counts (e.g. "12 of 200 today"). Private per-person sessions; What-it-remembers deletable.
atlas:TEAM-02: Four org roles with a permission table, invite links, member removal, per-device sign-out, and plan-scoped surfaces (ROI, meetings). No enterprise SSO, SCIM, or unauthorized-access testing evidenced. cross-checked
atlas:TEAM-02
Four org roles with a permission table, invite links, member removal, per-device sign-out, and plan-scoped surfaces (ROI, meetings). No enterprise SSO, SCIM, or unauthorized-access testing evidenced.
- No SSO/SCIM; revocation is membership/device/session level; no access testing.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Voice (audio, members, every plan) and video meetings (members + guest links, recording/transcript/AI note; plan-gated) started from any conversation, run in the web app; voice-only fallback when plan lacks meetings.
-
Finding: Web app ROI: spend vs commits/active time/lines changed by project/developer/model/session; Admin/Product Owner/owner only; covers readable projects; deleted sessions leave the figures.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
-
Finding: Bug reports need Atlas version/OS, agent/model, session (exported file or share link) and logs; cloud sessions shareable with Anyone public links (password, stop sharing); session may contain file contents/command output with secrets redacted pre-storage.
-
Finding: Join flow: install, create account, request at credits.tryatlas.cc (beta tester option), accept atlas-beta invite. "Atlas team reviews every sign-up"; shared cloud sessions visible to all members — keep private code in own org.
atlas:TEAM-03: Four permission modes with project-confined sandbox and network gating; command/tool/outward approvals with Allow-once/Deny; push stays manual and merges double-confirm. No approval lifetimes, tunnels, or secret-access approvals documented. cross-checked
atlas:TEAM-03
Four permission modes with project-confined sandbox and network gating; command/tool/outward approvals with Allow-once/Deny; push stays manual and merges double-confirm. No approval lifetimes, tunnels, or secret-access approvals documented.
- No approval lifetimes, tunnels, or secret-access approvals.
-
Finding: Runs Atlas Agent plus any ACP registry agent (Claude Code, Codex, Cursor, OpenCode and more) in the same chat; Atlas supplies the project. Atlas Agent permission modes Ask/Accept edits/Plan/Bypass (Shift-Tab cycles); in every mode except Bypass, commands run in a sandbox writable only inside the project with no network access unless approved. Model list comes from the Atlas gateway. Atlas Agent takes text prompts only, no images yet.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: 0.4.1 unreleased/in development (remote control, Ask Atlas, appearance). 0.4.0 unreleased (notifications pipeline with Allow once/Deny banners, keymap presets, agent switching keeps conversation, /remember, AGENTS.md mirroring, Cheaper Inference BYOK). 0.3.4 2026-09-27 Teams Early (cloud timeline, comments, org Atlas Agent, Linux, OrcaRouter BYOK). 0.3.3 2026-09-19, 0.3.2 2026-09-16 (first Windows build), 0.3.1 2026-09-06 (comms alpha, Atlas Agent billing, slash commands, config.toml), 0.3.0 2026-08-25 (ACP + timeline). Full history on GitHub.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
atlas:TEAM-04: Timeline and PR histories attribute sessions, reviews, comments, edits and merges across users and sessions, with exports; deletions and retention windows bound the record. No append-only audit model claimed. cross-checked
atlas:TEAM-04
Timeline and PR histories attribute sessions, reviews, comments, edits and merges across users and sessions, with exports; deletions and retention windows bound the record. No append-only audit model claimed.
- No audit-grade integrity model; history is editable/deletable by design.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Policy tab reads CLAUDE.md, AGENTS.md and Claude memory files into a table of 12 preference kinds with source text, file and match strength; edits rewrite the same span in the original file; Strong/Soft classification; Send a rule to agent chat.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Channels/DMs/group DMs with session/checkpoint cards; desktop notifications for DMs and mentions with per-kind switches; incoming webhooks give a channel a secret URL for CI/deploy/monitor posts (marked as integration). Needs sign-in + synced org.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:TEAM-05: Review-requested PR grouping, reading-order Guide tours, line-anchored comments with held reviews, stale-verdict marking, and merge-readiness sorting guide reviewers; tab state and drafts survive context switches. No timed fixture or missed-evidence measurement. cross-checked
atlas:TEAM-05
Review-requested PR grouping, reading-order Guide tours, line-anchored comments with held reviews, stale-verdict marking, and merge-readiness sorting guide reviewers; tab state and drafts survive context switches. No timed fixture or missed-evidence measurement.
- No measured fixture; narrow-screen and missed-evidence behavior unknown.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: app.tryatlas.cc: synced org in browser (same synced record; runs no coding agents). Google/GitHub sign-in; inbox with email notifications + quiet hours; dashboard (usage/activity); projects incl. shared threads view; Account holds access token for scripts, sign-in methods, active sessions incl. per-device sign-out.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Default shortcuts incl. Cmd-K palette, Cmd-T surfaces, Cmd-Shift-B source control, Cmd-J terminal, Alt-/ agent cycle, Shift-Tab mode cycle; keymap presets VS Code/Cursor/Zed/JetBrains with import/export; layouts; hint navigation. Recent Sessions shows per-session project, local-vs-worktree, agent, recency.
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:TEAM-06: Teammates take the driver seat of a live room with full transcript context; shared threads target same-checkout handover (desktop sharing in progress); memory briefings carry decisions across agents. Formal ownership transfer and post-disconnect runtime continuity unproven. cross-checked
atlas:TEAM-06
Teammates take the driver seat of a live room with full transcript context; shared threads target same-checkout handover (desktop sharing in progress); memory briefings carry decisions across agents. Formal ownership transfer and post-disconnect runtime continuity unproven.
- Runtime stays with the sharer; approvals stay on their desktop; formal transfer absent.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: "In progress: being built, and not in a release yet." Vision: shared checkout where teammates and agents type/run agents together with merged turns and resolved conflicts; thread view live in web app today (watch, resolve conflicts, compare/restore versions, comment, ask teammate's agent to run a prompt); sharing/typing/running/owner controls arriving in desktop.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
atlas:SHIP-01: Diffs in session review panels and PR Code view; call recordings/transcripts; session/memory/usage exports; timeline search retrieves post-session work with commit links. No dedicated artifact gallery or prompt-leak assessment. cross-checked
atlas:SHIP-01
Diffs in session review panels and PR Code view; call recordings/transcripts; session/memory/usage exports; timeline search retrieves post-session work with commit links. No dedicated artifact gallery or prompt-leak assessment.
- No artifact gallery; prompt-leak behavior unassessed.
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Voice (audio, members, every plan) and video meetings (members + guest links, recording/transcript/AI note; plan-gated) started from any conversation, run in the web app; voice-only fallback when plan lacks meetings.
-
Finding: Shared realtime canvas per conversation (notes/text/frames/shapes/images/video, undo); Atlas Agent can draw on it; conversation-scoped. Distinct from local project canvas (.atlas/canvas.json, not shared).
-
Finding: Markdown notes with blocks, backlinks, @ file mentions, icons/covers/properties; graph view; import folders (e.g. Obsidian vaults). Attach via @ in composer as visible context block; memory-searchable by agents.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Bug reports need Atlas version/OS, agent/model, session (exported file or share link) and logs; cloud sessions shareable with Anyone public links (password, stop sharing); session may contain file contents/command output with secrets redacted pre-storage.
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:SHIP-02: PR checks/required-checks gating with stale-verdict marking, human review flow, and agent Fix flows exist; Atlas never auto-pushes. No enforced test/lint gates, no spec-conformance checks, and no induced-failure proof. cross-checked
atlas:SHIP-02
PR checks/required-checks gating with stale-verdict marking, human review flow, and agent Fix flows exist; Atlas never auto-pushes. No enforced test/lint gates, no spec-conformance checks, and no induced-failure proof.
- Gates are surfaced, not enforced; no failure/conflict cases run.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
atlas:SHIP-03: Full GitHub PR workflow inside Atlas: branch/worktree checkout, Create PR, review threads, checks, rebase/merge conflict resolution, revert, branch delete — all over gh with Atlas never pushing for the user. cross-checked
atlas:SHIP-03
Full GitHub PR workflow inside Atlas: branch/worktree checkout, Create PR, review threads, checks, rebase/merge conflict resolution, revert, branch delete — all over gh with Atlas never pushing for the user.
- Push stays manual by design; authority ultimately GitHub-side.
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:SHIP-04: Merge authority surfaced in-product (method choice, auto-merge, double confirmations, revert) with reviews and stale-verdict invalidation; actual permission is the user’s GitHub access. No release/deployment authority exists. cross-checked
atlas:SHIP-04
Merge authority surfaced in-product (method choice, auto-merge, double confirmations, revert) with reviews and stale-verdict invalidation; actual permission is the user’s GitHub access. No release/deployment authority exists.
- Authority is GitHub-side; no release/deploy path.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
atlas:SHIP-05: Checkpoints bind tasks to commits/PRs with retained transcripts and guides; settle-on-merge tracks delivery; revert-PR and conversation undo cover rollback. Approval invalidation on new revisions is shown as stale verdicts only. cross-checked
atlas:SHIP-05
Checkpoints bind tasks to commits/PRs with retained transcripts and guides; settle-on-merge tracks delivery; revert-PR and conversation undo cover rollback. Approval invalidation on new revisions is shown as stale verdicts only.
- Delivery ends at merge; rollback is revert-PRs, not deployment rollback.
-
Finding: Built-in agent, no install, runs through the Atlas gateway billed to the organization; needs sign-in, synced org and AI access (composer bar explains what is missing). Reads/edits code, runs commands, skills, shared memory, navigates the Atlas window, acts in Cloud projects with per-action approval. Built-ins: /compact /diff /review /goal /init /status /undo.
-
Finding: Changes view: fetch/pull/push, branch switcher/create/delete/merge, stage/discard files and hunks, amend, conflict resolution (ours/theirs), revert/cherry-pick/tag. History shows "Produced by N sessions" per commit linking to the session. Auto-fetch from remote on by default (tracking branches only). Reference clones under .atlas/repos/.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Canonical definitions: project (.atlas/ dir, gitignored); workspace (org's shared record copy); session (one agent conversation, resumable); checkpoint (session↔commit link, survives amends/rebases, orphans on ambiguity); timeline; comment threads. Desktop finds checkpoints by watching refs without git hooks; CLI uses post-commit hook.
atlas:SHIP-06: PR list with Authored/Review-requested/Others groups, risk/size filters and sorts, bulk close, quick merge, retained guides/threads, and stale marking supports batch review; no six-PR fixture run and no dependency-aware merge queue. cross-checked
atlas:SHIP-06
PR list with Authored/Review-requested/Others groups, risk/size filters and sorts, bulk close, quick merge, retained guides/threads, and stale marking supports batch review; no six-PR fixture run and no dependency-aware merge queue.
- No batch fixture run; no dependency-aware merge/release queue.
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
atlas:OWN-01: Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished. cross-checked
atlas:OWN-01
Desktop app repository is Apache-2.0 (LICENSE + API + README badge) at the pinned commit; CLI ships from a separate public repo whose license was not verified; hosted org/cloud terms unpublished.
- CLI repo license and hosted-service terms unverified.
-
Finding: Owner tryatlas (org name Atlas), description "Source control for agents...", Apache-2.0 license, 9394 stars, default branch main, pushed 2026-11-11. Homepage https://www.tryatlas.cc/. Former pacifio/atlas path redirects here.
-
Finding: README: "Know which prompt shipped which line"; badges for CI, release, Apache-2.0 license, macOS/Windows/Linux, Discord, Atlas CLI, YouTube playlist; links to docs, contributing, issues.
-
Finding: alpha-0.4.1-org-agent published 2026-11-03; alpha-0.4.1 2026-10-19 (Atlas Teams Beta); alpha-0.3.4 2026-09-27; alpha-0.3.3 2026-09-19; alpha-0.3.2 2026-09-16.
-
Finding: Atlas CLI repo: "command-line version of Atlas... hooks into the coding agents you already use... captures every session live to the cloud... shared project memory"; language Go; public repo.
-
src-tauri/tauri.conf.json at pinned commit
Finding: Tauri app config: productName "Atlas", identifier "dev.atlas.ide", version 0.4.1, desktop window 1280x800 with overlay title bar; per-OS tauri.*.conf.json (macos/windows/linux) present.
-
Finding: Complete telemetry catalog: emitter src-tauri/src/telemetry/mod.rs, backend PostHog; toggle Settings → General → "Share usage data" defaults ON (opt-out, immediate); second toggle for account linkage; signed-in usage attributed to account (deliberate change in 0.2.4); source builds without a PostHog key are inert; never collected: prompts, code, file paths, keys (per file).
-
Finding: Repository root LICENSE is the Apache License 2.0 grant (matches API-reported Apache-2.0). CLI repository license was not verified in this pass.
atlas:OWN-02: Local-first: desktop/CLI run on user machines with source builds documented; session/memory/usage exports in MD/JSON/PDF; org leave/delete and CLI uninstall remove vendor state. No self-hosted control plane or whole-org export. cross-checked
atlas:OWN-02
Local-first: desktop/CLI run on user machines with source builds documented; session/memory/usage exports in MD/JSON/PDF; org leave/delete and CLI uninstall remove vendor state. No self-hosted control plane or whole-org export.
- No self-hosted org control plane; no whole-org portability.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
-
Finding: Markdown notes with blocks, backlinks, @ file mentions, icons/covers/properties; graph view; import folders (e.g. Obsidian vaults). Attach via @ in composer as visible context block; memory-searchable by agents.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
atlas:OWN-03: On-machine redaction before storage, US hosting, owner-controlled deletion, and a complete pinned telemetry catalog with opt-out toggles; source builds inert without a key. Training use, subprocessors beyond PostHog/models.dev, and full retention policy undisclosed. cross-checked
atlas:OWN-03
On-machine redaction before storage, US hosting, owner-controlled deletion, and a complete pinned telemetry catalog with opt-out toggles; source builds inert without a key. Training use, subprocessors beyond PostHog/models.dev, and full retention policy undisclosed.
- Training use, full subprocessor list, retention schedule, and first-run prompt undisclosed.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Per-project capture records prompts, responses, tool calls, file changes; Local (.atlas/ on machine, no account/network) or Cloud (same record synced to org). Redacts secrets on the machine before storing. Cloud needs git, sign-in, synced org. Commits become checkpoints surviving amends/rebases; squash ambiguity orphans rather than mislinks. Pause/resume; degraded-status dot with reasons and repairs. A folder needs no git repo (offers git init).
-
Finding: PRs in tabs beside sessions via gh (uses its sign-in; states missing/signed-out/rate-limited). List split Authored/Review requested/Others with state, checks, review state, search qualifiers, filters, sorts; quick Close/Merge; stack chains drawn indented. Merge button with method (Merge/Squash/Rebase), auto-merge, conflict resolution via agent session, revert, delete branch; second-click confirmations. Watched PRs checked every 2 minutes even with window hidden; check failures, passes, comments, conflicts message the session as "Pull request watch" rows; watch ends on merge/close/settle/stop/8 read failures. Guide tours, line comments, held reviews, stale-verdict marking. Settle-on-merge setting.
-
Finding: Per-repository shared record (.atlas/memory/, not synced to teammates) across worktrees/subdirectories; cross-project promotion in ~/.atlas/memory on repeat high-confidence content. Memory tool server handed to Atlas Agent and agents advertising HTTP MCP support. Kinds: plan, file changed, decision, fact, failure, architecture. Handoff summarizer (Raw/Provider), AGENTS.md mirroring, edit/forget/clear entries, memory verify via CLI. Code index with local embeddings; rebuild on model change.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Synced-org roles: Admin (invite, roles, links, rename, models, usage, ROI, sharing policy, delete org), Product Owner (ROI), Developer/Member (read timeline, sync, comment, chat/calls/drafts/Spaces). Creator is Owner: cannot be removed/leave, only one who permanently deletes sessions, ownership not transferable. Private channels/DMs unreadable even by admins.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: One agent per session; transcript incl. file reads/commands/edits; second session beside it (Cmd-T A); reopen repaints transcript from disk while agent reconnects. Sidebar lists sessions across org projects with project/branch/status, search, pins, snooze/settle (auto-settle idle 3 days, never running/waiting), per-branch worktree starts with checkout-lock checks. Import sessions agents kept on their own (incl. terminal).
-
Finding: Preferences in ~/.config/atlas/config.toml (commented, hand-editable, invalid edits keep last-good with error shown). Per-project: default agent, Atlas Agent permissions, code search, auto-fetch, git blame, memory jobs (off-for-all stays off). Privacy page: "Share usage data" toggle. Library: Agents (env vars incl. keychain-marked secrets, models incl. custom model by name, provider keys), Skills, Layouts.
-
Finding: Every config.toml key documented: auto-add .atlas to .gitignore, keepAwakeWhileRunning (prevent idle sleep while an agent works; macOS/Linux), confirm-before-delete/quit guards, per-kind notification switches (native banners + sounds; terminal command thresholds; rate-limit/retry warnings), model pricing source models.dev.
-
Finding: Multiplayer defined as shared record: timeline of every recorded session with prompts/tool calls/commits, comments beside the work, team chat/calls/drafts/Spaces/shared threads; organization owns sharing; cloud projects merge teammates' sessions; source code never syncs.
-
Finding: Bug reports need Atlas version/OS, agent/model, session (exported file or share link) and logs; cloud sessions shareable with Anyone public links (password, stop sharing); session may contain file contents/command output with secrets redacted pre-storage.
-
Finding: In-app feedback (Issue/Feature/Improve/Other) with screenshot attach, sent even with usage data off; anonymous option; public Discord and GitHub Bug Report / Feature Request forms.
-
Finding: Complete telemetry catalog: emitter src-tauri/src/telemetry/mod.rs, backend PostHog; toggle Settings → General → "Share usage data" defaults ON (opt-out, immediate); second toggle for account linkage; signed-in usage attributed to account (deliberate change in 0.2.4); source builds without a PostHog key are inert; never collected: prompts, code, file paths, keys (per file).
atlas:OWN-04: Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished. cross-checked
atlas:OWN-04
Public monthly prices: Free $0, Pro $20, Team $100, Business $200 with member/project/storage/AI/vectorization quotas; billing managed in the app (login-walled, not inspected). Overage rates, taxes, and commitment terms unpublished.
- Overages, taxes, commitment, and billing mechanics (login-walled page) unstated.
-
Atlas — Every coding agent, one workspace, your whole team
Finding: Native macOS workspace installing Claude Code, Codex, OpenCode and ACP agents across projects; every session recorded on an organisation timeline with team chat beside it.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: v0.4.1; macOS 13+ (Apple Silicon and Intel), Windows 10+ (x64); "Native binary for Apple Silicon, Intel and Windows, no Electron"; "The Atlas account is optional"; address Atlas 1111B S Governors Ave #39146 Dover, DE 19904; links to GitHub, X (@tryatlas_cc) and Discord.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Beta run inside Atlas itself (atlas-beta org team chat/calls/drafts); public Discord open; contributors path via Contributing.
atlas:OWN-05: Fixed ADE cost from public prices (e.g. 1 dev on Free $0; 2 devs on Free $0 within 3 seats; 5 devs on Pro $20/mo); included AI $8–$80/mo; agent subscriptions/keys and any overages variable and unpriced here. No per-task cost data exists (local execution, estimates only). cross-checked
atlas:OWN-05
Fixed ADE cost from public prices (e.g. 1 dev on Free $0; 2 devs on Free $0 within 3 seats; 5 devs on Pro $20/mo); included AI $8–$80/mo; agent subscriptions/keys and any overages variable and unpriced here. No per-task cost data exists (local execution, estimates only).
- No per-task metering; variable model/compute spend stays outside Atlas pricing.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Web app ROI: spend vs commits/active time/lines changed by project/developer/model/session; Admin/Product Owner/owner only; covers readable projects; deleted sessions leave the figures.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
atlas:OWN-06: Alpha 0.4.1 in active development (release 2026-11-03, main commit 2026-11-11); beta program with public Discord and GitHub issues; in-app feedback; Google/GitHub sign-in with AI access granted by request. No SLAs, status page, or funding disclosed. cross-checked
atlas:OWN-06
Alpha 0.4.1 in active development (release 2026-11-03, main commit 2026-11-11); beta program with public Discord and GitHub issues; in-app feedback; Google/GitHub sign-in with AI access granted by request. No SLAs, status page, or funding disclosed.
- No SLAs, status history, or deprecation policy; alpha software.
-
Finding: v0.4.1; macOS 13+ (Apple Silicon and Intel), Windows 10+ (x64); "Native binary for Apple Silicon, Intel and Windows, no Electron"; "The Atlas account is optional"; address Atlas 1111B S Governors Ave #39146 Dover, DE 19904; links to GitHub, X (@tryatlas_cc) and Discord.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: 0.4.1 unreleased/in development (remote control, Ask Atlas, appearance). 0.4.0 unreleased (notifications pipeline with Allow once/Deny banners, keymap presets, agent switching keeps conversation, /remember, AGENTS.md mirroring, Cheaper Inference BYOK). 0.3.4 2026-09-27 Teams Early (cloud timeline, comments, org Atlas Agent, Linux, OrcaRouter BYOK). 0.3.3 2026-09-19, 0.3.2 2026-09-16 (first Windows build), 0.3.1 2026-09-06 (comms alpha, Atlas Agent billing, slash commands, config.toml), 0.3.0 2026-08-25 (ACP + timeline). Full history on GitHub.
-
Finding: Commands: login/logout/whoami/org, init/settings/status, agents add/remove/list/sessions, capture enable/disable/sync/retry/link, sync on/off/serve, memory browse/list/index/enable/disable/export/import/verify, update/uninstall. Piped output is plain text; flags answer prompts. Agents themselves call hidden atlas-cli hooks and atlas-cli mcp-bridge.
-
Finding: Local Personal org (no account, no network) vs synced org (members, cloud projects, chat, Atlas Agent). Cloud handle unique; region picker: US hosted, EU not available yet. Create/switch/rename/delete orgs from the menu; deleting removes its projects and sessions but never project files.
-
Finding: Board of every recorded session newest-first grouped by day with live/needs-attention/finished/imported marks, model/branch/tokens/checkpoints, 15s org refresh, newest 500 sessions, author filter. Comments anchored to session/message/tool call/checkpoint, threaded, resolvable. Export session as Markdown/JSON; share links (cloud) incl. public links with password. Secrets redacted on machine before storing; pre-share secret counts shown. Timeline questions use own API key (Settings → Agents → Provider keys).
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Per-machine usage: tokens/messages/sessions/estimated cost over 7/30/90d or custom, daily chart grouped by Project/Agent/Model, efficiency figures (cache hit rate, ratios), searchable tables; token detail only when ACP implementation reports usage; cost via models.dev USD prices, no invented fallback. Export PDF/JPEG/Markdown. Org Atlas Agent spend per person under AI budgets in web app.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
-
Finding: Beta run inside Atlas itself (atlas-beta org team chat/calls/drafts); public Discord open; contributors path via Contributing.
-
Finding: In-app feedback (Issue/Feature/Improve/Other) with screenshot attach, sent even with usage data off; anonymous option; public Discord and GitHub Bug Report / Feature Request forms.
-
Finding: Join flow: install, create account, request at credits.tryatlas.cc (beta tester option), accept atlas-beta invite. "Atlas team reviews every sign-up"; shared cloud sessions visible to all members — keep private code in own org.
-
Finding: Owner tryatlas (org name Atlas), description "Source control for agents...", Apache-2.0 license, 9394 stars, default branch main, pushed 2026-11-11. Homepage https://www.tryatlas.cc/. Former pacifio/atlas path redirects here.
-
Finding: alpha-0.4.1-org-agent published 2026-11-03; alpha-0.4.1 2026-10-19 (Atlas Teams Beta); alpha-0.3.4 2026-09-27; alpha-0.3.3 2026-09-19; alpha-0.3.2 2026-09-16.
atlas:OWN-08: Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated. cross-checked
atlas:OWN-08
Free $0 plan (3 members, Standard storage/meetings, shared threads) plus a free Apache-2.0 app download; local work needs no account. Two-person collaboration fits Free with own agent subs; AI allowance, meetings recording, and Remote Runs are paid-gated.
- Paid gates on AI allowance, meeting recordings, Remote Runs; billing page uninspected.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.
-
Finding: v0.4.1; macOS 13+ (Apple Silicon and Intel), Windows 10+ (x64); "Native binary for Apple Silicon, Intel and Windows, no Electron"; "The Atlas account is optional"; address Atlas 1111B S Governors Ave #39146 Dover, DE 19904; links to GitHub, X (@tryatlas_cc) and Discord.
-
Finding: Desktop app for macOS 11+ (Apple silicon and Intel .dmg), Windows 10+ x64 (.msi, WebView2 fetched by installer), Linux x86-64 (.deb, .rpm, Arch, AppImage, tarball; needs WebKitGTK 4.1 and bubblewrap). macOS installs an `atlas` command in ~/.local/bin; Linux installs `atl`/`tryatlas`; "The atlas command is not available on Windows yet". macOS/Windows auto-update in background. Atlas 0.4.1; no account needed.
-
Finding: Fresh install to first session: install Atlas, add a project (New/Local folder/Git URL/GitHub owner-repo via gh), add an agent (Atlas Agent needs synced org with AI access; otherwise install from ACP registry), turn on capture; second half connects the team via synced organization.
-
Finding: A live desktop session can be put in a room opened from any browser (web app sidebar lists rooms) or a phone browser ("built for a phone screen"; not a native phone app). Room carries prompts, Stop, model/mode changes; agent keeps running on the desktop with its files, credentials and spend. One driver at a time; invites need an organization. Permission prompts stay "Waiting for approval on the desktop"; bypass/full-access modes are Desktop only.
-
Finding: atlas-cli hooks terminal agents (Claude Code, Codex, Cursor, Copilot CLI, OpenCode, Factory Droid, Antigravity, Pi): records sessions, links commits, syncs to org timeline, one shared memory. macOS (Apple silicon/Intel) and Linux (x86-64/arm64); not Windows. curl installer with checksum verification into ~/.local/bin. Device-code sign-in incl. --no-browser for SSH. Same sign-in file and .atlas/ format as desktop.
-
Finding: Voice (audio, members, every plan) and video meetings (members + guest links, recording/transcript/AI note; plan-gated) started from any conversation, run in the web app; voice-only fallback when plan lacks meetings.
-
Finding: Atlas Agent billed per request to the org via gateway; needs synced org + granted AI access (plan: monthly + optional daily budget, covered models; off until a plan exists; request at credits.tryatlas.cc). Other agents run on own subscription/keys. Admins narrow models within the plan; gateway refuses disabled models. Caps enforced: org stops new requests at cap; per-member caps; over-quota Atlas Agent behavior. Prompt capture windows (hours-to-day) for troubleshooting only.
-
Finding: Editor/terminal/git/ACP agents/memory/local capture/personal timeline work signed out in a local org. Sign-in unlocks synced org, Atlas Agent (+AI access), cloud projects, chat/calls/drafts/Spaces, web app. Google or GitHub sign-in (connectable both), device-code flow, per-device Active sessions with remote sign-out. Access token section: short-lived personal token for scripts.
-
Finding: Join flow: install, create account, request at credits.tryatlas.cc (beta tester option), accept atlas-beta invite. "Atlas team reviews every sign-up"; shared cloud sessions visible to all members — keep private code in own org.
-
Finding: Owner tryatlas (org name Atlas), description "Source control for agents...", Apache-2.0 license, 9394 stars, default branch main, pushed 2026-11-11. Homepage https://www.tryatlas.cc/. Former pacifio/atlas path redirects here.
-
Finding: README: "Know which prompt shipped which line"; badges for CI, release, Apache-2.0 license, macOS/Windows/Linux, Discord, Atlas CLI, YouTube playlist; links to docs, contributing, issues.
-
Finding: Complete telemetry catalog: emitter src-tauri/src/telemetry/mod.rs, backend PostHog; toggle Settings → General → "Share usage data" defaults ON (opt-out, immediate); second toggle for account linkage; signed-in usage attributed to account (deliberate change in 0.2.4); source builds without a PostHog key are inert; never collected: prompts, code, file paths, keys (per file).
atlas:OWN-09: Units, included limits and plan prices are public ($/mo, members, projects, GB, vectorization tokens, meeting tier); overage rates, model/compute pass-through, minimum commitment and cancellation terms are unstated. Partially specified, not quote-only. cross-checked
atlas:OWN-09
Units, included limits and plan prices are public ($/mo, members, projects, GB, vectorization tokens, meeting tier); overage rates, model/compute pass-through, minimum commitment and cancellation terms are unstated. Partially specified, not quote-only.
- Overages, pass-through, commitment and cancellation terms missing.
-
Finding: Plans compared: Free $0/month, Pro $20/month, Team $100/month, Business $200/month. Members 3/5/25/100; synced projects Unlimited/10/50/Unlimited; session storage Standard/10GB/100GB/500GB; AI each month Not included/$8/$40/$80; vectorization tokens Unlimited/20M/100M/300M; concurrent runs per thread 4/4/4/10; Remote Runs excluded on Free/Pro.