November 2026 status archive, page 11 of 15
Shutdown notices, recent changes and all monthly archives
These events link to today's eligible findings, not historical before-and-after claims.
Cited events
20 ADEs
gh-awinitial researchresearch correction29 findings · 17 sources
Findings today
- gh aw ships a native GitHub CLI extension for macOS, Windows and Linux; this is CLI/TUI access, not a separate…
- Built-in engines and custom imported definitions are separate.
- A personal Copilot subscription can use COPILOT_GITHUB_TOKEN.
- Install the extension, select/authenticate an engine, add Markdown workflow source and compile its lock file.
- gh aw compile, logs and audit provide workflow authoring and inspection operations.
- Declared MCP servers are filtered through the configured gateway tool allowlist.
- Workflow Markdown holds natural-language task instructions; YAML frontmatter sets triggers and policy.
- Cache-memory persists declared files between runs with workflow scope/retention.
- Workflows on distinct issues/PRs/branches may run concurrently; default per-engine groups serialize each engine.
- queue:max and cancel-in-progress settings determine pending-run retention and obsolete-PR cancellation.
- Schedule frontmatter can run a recurring workflow such as a daily report.
- Compiled workflows execute on GitHub Actions runners, including user-controlled Linux runners.
- AWF Docker isolation is the default; opt-in Cloud Hypervisor/NVX preview runtimes use KVM microVMs on eligible runners.
- Experimental engine.durable saves/restores native session checkpoints in artifacts, default seven days subject to…
- Top-level max-turns defaults to 500 for supported engines; Agy lacks this limit.
- gh aw logs and audit expose per-run duration, token/credit usage and tool/session information.
- AI Credits estimate inference per workflow run; Actions compute is a separate cost component.
- Workflow and engine concurrency groups bound overlap and preserve queued work where configured.
- The gateway applies explicit tool allowlists; engine-level tool enforcement differs by engine.
- The extension exposes workflow logs, audit reports and output artifacts, not a product UI for reviewing proposed…
- Workflow configuration separates deterministic CI jobs from agent reasoning and permitted outputs.
- create-pull-request safe outputs propose code in a separate permission-scoped job.
- Merge operations are separately configured safe outputs with scoped credentials; PR creation alone does not grant…
- The gh-aw compiler and runtime repository use MIT.
- The product code and execution worker can run on user-controlled runners, but GitHub Actions supplies workflow…
- The gh-aw extension/compiler is free MIT software without a paid extension plan.
- Release v1.2.1 dated 2026-11-06 adds experimental artifact-backed continuation and workflow fixes.
- gh-aw itself is free; model and runner charges remain separate.
- The application adds no platform charge; inference and Actions minutes remain separate billed units.
Sources
Open the gh-aw page
Ghostexinitial researchresearch correctionfreshness recheck34 findings · 5 sources
Findings today
- The product ships desktop/native phone clients.
- Multiple coding agents are launched and represented in chat or terminal views.
- Users bring their existing agent; usage/account switching is documented for Claude and Codex.
- Installers and existing agent setup are documented; no first-task timing was measured.
- The ghostex command lets agents open sessions, prompt other agents and read replies.
- Optional views and plugins extend the workspace; provider tools remain separate.
- Vendor identifies a Rust/GPUI desktop shell with native terminal and Chromium/VS Code panes.
- Projects organize sessions and optional Kanban work.
- Plans can be opened, annotated and returned to an agent with line-number feedback.
- Session history and reference files can be revisited, without a verified managed memory contract.
- The optional integrated Beads board provides persistent issue cards with state and metadata separate from agent…
- Multiple sessions and coordinator threads are documented; concurrency quotas remain unknown.
- Ghostex coordinates child sessions through product commands.
- Scheduled prompts are listed as a product automation capability.
- Follow-ups may be queued or delivered to interrupt current work.
- Sidebar threads and coordinator views expose multiple sessions, without a tested fleet fixture.
- Agent/model/effort can be selected and a coordinator can manage multiple threads.
- Ghostex runs agents locally, on SSH machines or in boxes created in the user external cloud-provider account.
- Worktrees and optional agentbox Docker boxes are documented.
- Past sessions can be searched and resumed; crash recovery was not tested.
- Usage limits and session progress are visible; totals and failures were not reconciled.
- Shared chat guests can send instructions under host-controlled modes.
- Shared chat links have message controls, codes, expiry and immediate revocation.
- Shared chat activity has a guest message log.
- Readable diffs and annotations support review and feedback.
- Files, plans, HTML and diagrams can remain beside the conversation.
- Changes view integrates branches, worktrees, commits and pull requests.
- Changes view offers hunk/file revert and PR state links.
- Ghostex repository uses MIT; bundled third-party components retain separate licenses.
- Local desktop code is available and users connect their own remote machines.
- Settings document an analytics toggle and its default.
- Application is free; no per-task provider cost was established.
- Active release artifacts and community support are public, with iOS TestFlight access restrictions.
- Vendor explicitly says Ghostex is free; provider and compute costs are separate.
Sources
Open the Ghostex page
Happyinitial researchresearch correction37 findings · 40 sources
Findings today
- Happy provides Desktop for macOS, Windows and Linux; mobile iOS, Android and web; and its original happy CLI…
- Happy Agent runs its own loop by default and calls Claude, Codex and Grok models on existing CLI sign-ins.
- Happy reuses existing Claude Code, Codex and Grok CLI sign-ins and their subscriptions; Antigravity and ACP agents…
- Importing registers an existing folder without moving, copying or initializing Git.
- Happy Agent exposes a token-authenticated REST API for sessions, streaming events and control on user infrastructure.
- Skills are SKILL.md files and MCP servers are mcp.toml entries; agents can write both.
- The built-in Chief of Staff handles setup and task creation through ordinary conversation, including projects,…
- Subagent trees are capped at depth three and at most five agents including finished children.
- AGENTS.md files in the project tree are read by each agent and reread when they change; global AGENTS.md applies…
- Subagents run in the background by default and return completion to the parent.
- Bots and their subtasks may create a full conversation in a new workspace, with the creator as coordinator.
- Agents delegate bounded work, choose each child model and effort, receive background completion notices and send…
- Primary sessions can schedule messages to known agent IDs, list or cancel them and survive intervening runtime restarts.
- Agent questions remain in the conversation and experimental Inbox; answering on one device closes the shared state.
- Per-session UI shows phase, tools, background processes, subagents, waiting state, whole-tree token usage and…
- The built-in Chief of Staff configures accounts, imports projects, pairs phones, deploys remote agents and creates…
- Happy Agent can run on an always-on user server while Desktop connects over Tailcat and phones connect through an…
- Parallel workspaces use Git worktrees on branches cut from the remote trunk.
- Conversations and child sessions persist across window closure and runtime restart; finished children accept…
- New workspaces copy configured sync files and run setup_commands.
- Smart provider pools can rotate same-provider accounts after authentication or quota-exhaustion errors.
- Session status includes active tools, background work, subagents, waiting state, tokens and changed files.
- Provider quotas are polled about every fifteen minutes and shown in the app; they are observations, not billing records.
- Team members see the same projects and sessions and can speak to the agent.
- Team mode verifies per-user WorkOS tokens.
- Read only, Workspace write, Auto and Full access share a permission model.
- Session transcripts retain provider-specific opaque data; compaction creates a visible message describing replaced…
- The desktop interface has conversations, files, diffs and an attachable terminal; current docs describe changed…
- Named teammates operate on shared sessions and projects on the same team server.
- Happy Desktop shows files and diffs with conversations, and its documentation includes a recipe for a live, shared…
- The mobile app, original CLI and relay repository has the MIT license.
- The core Happy Agent runs self-hosted on user machines or servers, with Desktop connecting remotely.
- Happy apps send pseudonymous usage events to PostHog, excluding message content, prompts, code and file names, and…
- Happy is free and MIT licensed.
- Happy Agent v0.4.108 adds GPT-6.1 Terra and stop-state persistence; Desktop v0.0.102 has opt-in beta Stem cells.
- The desktop application and self-hosted runtime are free.
- The core ADE has no software subscription charge; coding-model subscriptions, host resources and optional voice are…
Sources
Open the Happy page
Herdrinitial researchresearch correction31 findings · 34 sources
Findings today
- Herdr TUI runs as a Rust terminal program with a product CLI on macOS, Linux and Windows; it is not an Electron…
- Herdr TUI runs existing agent CLIs in real terminal panes.
- Install the Herdr binary or a package-manager build, start it where the repository lives and run existing agent…
- Herdr provides CLI wrappers and a local socket API for workspaces, tabs, panes, agent prompts, waits, state events…
- Plugins extend Herdr terminal panes and workflows, with a marketplace and CLI/API management.
- Herdr TUI is a Rust binary running in the user terminal rather than an Electron app.
- Scripts and agents create and name workspaces, tabs and panes.
- Saved session shape retains workspaces, tabs, panes, current directories, layout, focus and supported native agent…
- Multiple agent terminals run concurrently; working, blocked, done and idle states roll up into tabs and workspaces.
- Agents and scripts can start, prompt and wait on selected panes.
- Herdr ships an agent skill and CLI that let an agent in a Herdr pane start agents in other panes, prompt them, read…
- Known agents report or are detected as idle, working or blocked, with notifications when they finish or need input.
- The sidebar rolls a blocked pane up to its tab and workspace, and shows active work and done agents until viewed.
- Users can start one terminal task, split panes, prompt agents and wait on their state through the same CLI/socket…
- Local and saved SSH machines each own their server, sessions and processes.
- The TUI socket exposes worktree.list, worktree.create, worktree.open and worktree.remove with workspace provenance.
- Detach keeps original processes alive.
- Restored panes return to saved directories or remain visibly errored if the directory or shell is unavailable.
- Agent waits have explicit until and timeout_ms controls; an atomic prompt-and-wait avoids the race between separate…
- State derives from screen manifests or integration reports.
- Remote connection authentication stays with OpenSSH.
- Optional pane history stores recent terminal output locally and may contain secrets.
- Herdr native app ships a Changes panel for reviewing Git file changes and diffs.
- Herdr TUI lets one client own a terminal at a time; herdr agent attach and terminal attach accept --takeover to…
- Optional local pane-history replay preserves recent terminal contents after restart.
- Herdr TUI source is distributed under Apache License 2.0 at the pinned repository revision.
- The full TUI and its server can be built from source and run on the user machine or SSH host.
- Session data and optional pane history are stored in the local Herdr state directory; history may contain prompts or…
- The vendor states the native app, herdr-core and Herdr TUI are free and open source.
- The October 14 release fixes prefix shortcuts, Codex integration warnings, blocked Claude questions and Windows issues.
- The app, core and TUI are offered free and open source.
Sources
Open the Herdr page
Jcodeinitial researchresearch correction40 findings · 43 sources
Findings today
- Jcode provides a terminal UI and product CLI on macOS, Linux and Windows.
- Jcode is its own coding harness with Claude, OpenAI, Gemini, Copilot and Azure provider logins, local Ollama/LM…
- Jcode supports subscription OAuth and direct API providers, Claude subscription login and OpenAI/ChatGPT/Codex…
- Install the binary, run jcode in the project folder and choose a provider login.
- The TypeScript SDK launches a private Jcode process or attaches through api-bridge to a local or forwarded remote…
- Jcode loads global and project MCP configuration and imports Claude Code/Codex MCP servers.
- Users start an interactive project session or submit a noninteractive prompt through jcode run.
- Plans consist of phases, lanes and items with names, descriptions, lifecycle status, priority and confidence.
- Phases are ordered, lanes can run in parallel and items run in lane order.
- Repository and global AGENTS.md load into the corresponding scope.
- When one agent edits a file another has read, the server notifies the reader, which can inspect the diff.
- Native saved todo items have stable internal IDs, names, descriptions, priority, confidence, states and phase/lane…
- The coordinator owns the shared versioned plan and can propose, approve, assign and control tasks.
- Multiple agents can work concurrently in the same repository; the server manages messaging, completion status and…
- Only root sessions spawn swarm workers.
- Jcode agents can autonomously spawn a swarm, making the original agent coordinator.
- Ambient mode is implemented but disabled by default.
- Enter interleaves a message with the working agent and Shift+Enter queues it.
- Desktop offers a spatial session canvas and zoomed-out overview with task and orchestration views.
- Users start quick prompts, headless requests or swarm sessions, with provider/model/effort and feature configuration.
- The daemon owns sessions, tools and provider calls.
- Swarm coordination supports optional Git worktrees and worktree manager roles for separate code work and later…
- Normal uninstall retains config, authentication, memory and sessions unless purge is requested.
- Cloud handoff does not make machines identical.
- cloud wait exits unsuccessfully on timeout without cancelling the remote turn.
- Todo lanes show worker activity, observed tokens, throughput, connection, elapsed time and tool intent.
- The /usage and /info commands show provider limits, session information and tokens.
- Ambient mode admits only one ambient instance and gives interactive sessions priority.
- Temporary local environment access is scoped to the handoff lease, defaults to twenty-four hours and ends on…
- Plan item identity is preserved through matching-name replacement and explicit rename.
- Jcode Desktop renders file diffs beside live sessions, and the TUI /diff command cycles the diff display mode.
- The desktop client renders diffs, Mermaid diagrams, inline images and tool cards beside agent conversations.
- The inspected public Jcode CLI/desktop implementation has no dedicated native PR creation or management operation;…
- Jcode source is MIT licensed at the pinned revision, including a free-of-charge permission grant.
- The Jcode daemon, TUI and tools can run on a user-controlled machine or SSH host, with direct model credentials or…
- Default usage telemetry may include provider OAuth email, installation ID and numeric/categorical usage.
- Optional hosted plans start at USD 10 monthly with USD 20 inference credit and fifty Jcode Cloud machine hours.
- v1.0.0-preview.199, published November 6, is an opt-in v1 pre-release of experimental binaries, with production use…
- The application can be downloaded, used and self-hosted as free MIT software.
- The hosted plans disclose recurring USD charges, inference credit, cloud hours, API-price overage and separate…
Sources
Open the Jcode page
loki-modeinitial researchresearch correctionfreshness recheck40 findings · 14 sources
Findings today
- CLI runs on macOS/Linux, with current Control Plane web UI and optional editor integrations.
- Current supported coding providers are Claude Code/Agent SDK and OpenCode, with model routes separate.
- Existing Claude subscription login and API-key/local-model routes are documented separately.
- Install and setup/doctor check provider and project prerequisites; offline tour needs no provider.
- CLI and current Control Plane API create/inspect/control runs and outputs.
- MCP and agent instructions extend the CLI; legacy API details are scoped separately.
- Task text, PRD files and external tracker issues are accepted; tracker identity remains external.
- Run derives a delivery contract and checks from input task before implementation.
- Workspace repo dependencies and layered integration preserve cross-repo work identity.
- Local memory and context are available; managed persistent spec conformance needs separate evidence.
- Workspace runs execute independent repos concurrently within configured cap.
- Workspace scheduler orders child repo runs; agent assignment/claim enforcement remains untested.
- Loki owns multi-repo child execution and aggregates signed results, with review council in each supported run.
- Official user-installed GitHub workflow can start issue work from labels/comments.
- Newest release can continue an ambiguous task with a disclosed choice capped at PARTIAL; generic README stop-and-ask…
- Current UI groups run outcomes/receipts/costs, without a verified 12-state fleet fixture.
- Small edits, interactive chat and autonomous task/PRD flows are distinct entry routes.
- Local CLI or user-managed runner hosts execution; detach alone does not survive laptop sleep.
- Workspace tasks use separate worktrees; optional Docker scopes filesystem and credentials.
- Local run records persist and detached tasks survive terminal closure; runtime continuity was not stress-tested.
- Run spend caps and outcome limits are explicit; newest release may continue ambiguous work as PARTIAL.
- Run outcomes and sealed receipt verdicts have distinct vocabularies.
- Receipts record measured session costs/tokens with null/unrecorded fields; subscription cost is not zero.
- Workspace concurrency and integration stage gates bound multi-repo scheduling.
- Current Control Plane does not expose legacy RBAC or OIDC login.
- Current Control Plane uses bearer auth and loopback-only machine-mutating routes.
- Signed receipt and event digest provide scoped integrity evidence, not proof of code correctness.
- PR descriptions summarize requested change, checks and not-proven items.
- Signed receipt and local artifacts retain run checks and results.
- Task-derived checks, fail-to-pass behavior and review council produce scoped receipt verdicts.
- Primary flow opens PR after building and verifying, with local branch/no-PR alternatives.
- Deployment CLI has a signed-receipt gate distinct from PR publication.
- Verification checks code drift and rollback support records post-action health.
- Current Control Plane exposes merge queue and review-risk endpoints; high-volume UI behavior was not tested.
- Current source is BUSL-1.1 with use restrictions; future Apache conversion is not current license.
- Current core and Control Plane self-host without a Loki account; provider services remain separate.
- Diagnostics defaults and opt-out controls are explicitly disclosed.
- Existing self-hosted core is free; model and compute charges are separate.
- Current version is publicly released with changelog and issue support.
- Free self-hosted core and offline receipt sample support evaluation.
Sources
Open the loki-mode page
Lovableinitial researchresearch correction38 findings · 40 sources
Findings today
- Lovable provides native iOS and Android apps using the same account, projects and credits as its web app.
- A workspace subscription pools build credits for its projects and collaborators.
- Users describe a project or change in chat; the hosted builder explores the code, edits files and runs checks…
- The official MCP server lets external AI tools interact with Lovable projects using authenticated tools.
- Lovable can connect to custom MCP servers so the agent can access external tools and information.
- Natural-language prompts initiate project changes in the project chat.
- Plan mode investigates without changing code and produces a plan users can edit, revise and approve before Build mode.
- Project knowledge provides persistent instructions and context for Lovable across project requests.
- Read-only subagents can search files, inspect project code, perform lookups and browse the web, returning findings…
- Lovable can run multiple read-only research subagents concurrently on independent questions.
- Lovable starts Generic and Explore read-only subagents, displays their status and inspected steps, and receives…
- Users can send steering follow-ups while a build runs or stop it.
- The draft switcher shows working, updates available, new changes, needs input and failed states, with shared All…
- A /goal Build message continues toward an outcome for up to ten hours, accepts follow-ups and still pauses for…
- Web and mobile clients operate on the same hosted projects and require an active network connection.
- Each draft is a separate project copy with its own chat and preview.
- Accepted drafts enter project history.
- Build messages run up to ten hours and wrap up during the last half hour.
- Details view has a Timeline of tool calls and steps and a Changes tab showing resulting file edits.
- The response menu shows Credits used during execution and final cost after completion.
- Distinct Lovable accounts invited as project Editors can build in the shared project and access its editor/chat.
- Project and workspace roles include Viewer, Editor, Admin and Owner with explicit edit, publish, member and billing…
- Editors and above can publish by default.
- The Details Timeline records build steps and tool calls and Changes shows edits.
- Users can inspect per-change summaries and file diffs through Details, then return to Preview.
- Build results include file diffs and summaries in Details, with Timeline and Changes tabs.
- Lovable has browser testing, frontend tests and edge-function verification; most verification tools run when requested.
- GitHub Cloud and Enterprise connections provide two-way code sync, branch operations and pull-request creation.
- Editors and above can publish, while Enterprise policies can require an approval or block publishing.
- Every manual code save creates a version in project history and users can revert a broken edit.
- Vendor-authored service terms grant limited access and reserve ownership of the platform and underlying software to…
- Paid-plan editors can export the project as a ZIP; Git sync is available on all plans.
- Business and Enterprise workspace data are excluded from AI training by default; Free and Pro can opt out per account.
- Free includes five daily build credits capped at thirty monthly.
- Current release notes describe new AI image models on November 6 and memory maintenance on November 4.
- Lovable publicly asserts SOC 2 Type II and ISO 27001 compliance.
- An ongoing Free plan provides daily build credits capped monthly, unlimited workspace members and Git sync.
- Public USD monthly and annual credit tiers and included allowances are documented; Build mode consumption is…
Sources
Open the Lovable page
MonoCodeinitial research24 findings · 9 sources
Findings today
- Desktop installers support macOS Apple Silicon/Intel, Linux x64 and Windows x64.
- Claude Code, Codex, Cursor, Grok Build, OpenCode, Pi, omp, fx, Hermes, Antigravity and Devin CLI are listed.
- README explicitly says existing supported coding-agent subscriptions work; MonoCode does not sell tokens.
- Install/authenticate a provider, open a project folder, select an agent and send a message.
- Opt-in app CLI offers models.list, sessions.start/list/read/send/stop and note/worktree operations.
- Production uses React/Tauri; experimental native preview uses Rust/GPUI.
- Jira inbox includes issue descriptions/comments and starts work with a Jira identifier; it is an external tracker…
- Remote Plan mode creates a reviewable plan card whose Build action continues on the host.
- Operator notes.read/write stores project notes with IDs, titles and tags.
- README explicitly documents separate Git worktrees for parallel tasks and split sessions.
- sessions.start accepts provider, model, effort, permission mode and worktree scope.
- Experimental Monos can run user-configured scheduled tasks.
- sessions.stop cancels the current local turn and pauses queued messages; archive preserves its conversation.
- Installed provider CLIs execute in the chosen local project or worktree.
- Desktop/operator can create and choose Git worktrees for session file separation.
- Remote host restart retains transcripts and marks active turns interrupted; uncertain provider operations are not…
- Matching host archives are checksum verified; missing release artifacts cause an explicit error instead of falling…
- Provider permissions remain CLI-specific; native Supervised mode does not guarantee approval on every edit.
- Remote host stores transcripts/snapshots and preserves pending request identity across desktop restarts.
- Desktop and remote Changes/Git views display file diffs and history.
- Remote Git views stage/discard files, commit and push through the host.
- MonoCode source is MIT licensed; provider trademarks and notices remain separate.
- Desktop and remote host can run from repository source on user-controlled machines.
- The React/Tauri app is production; Rust/GPUI remains experimental preview and uses the same saved database.
Sources
Open the MonoCode page
multi-agent-shoguninitial researchfreshness recheck31 findings · 3 sources
Nimbalystinitial researchresearch correctionfreshness recheck41 findings · 7 sources
Findings today
- Desktop runs on macOS/Windows/Linux; native iOS and Android companions and team browser documents/trackers have…
- Multiple provider lanes include Claude Code SDK/CLI, Codex SDK/app-server/ACP, Cursor Agent, Grok Build and…
- Provider lanes use entered keys or previously authenticated user CLIs; local LM Studio lane is separate from coding…
- Open/create document, work with agent and review edits.
- nim CLI queries workspace/session/document/tracker data, mutates trackers and offers local stdio wiki MCP.
- MCP client, internal tool servers and extension SDK provide custom editor/provider integrations.
- Desktop is Electron/TypeScript; iOS uses SwiftUI and Android Kotlin/Compose.
- Native typed tracker intake supports local creation, GitHub issue adoption and session links.
- Plans are native tracker types, linked to sessions/files/commits, with editable document bodies.
- Tracker ready view exposes unblocked work, dependencies and collections.
- Project knowledge memory, wiki pages and linked session context support persistent knowledge.
- Content search, code excerpts and browser AI tools supply code/context awareness.
- Product-owned tracker has typed bugs/tasks/plans, saved views, priorities, labels, assignments and durable states…
- Agents use tracker CRUD/readiness and session links; humans see the same persistent records.
- Plans and requirements can remain in linked tracker/document files with revisions and review states.
- Parallel sessions and Blitz can run across isolated worktrees.
- Coordinating sessions can spawn workers with explicit reporting policy and selected reasoning effort.
- Product-owned session tools spawn, inspect, message and receive results from child/sibling/isolated sessions.
- Markdown frontmatter automations schedule recurring AI work by interval, daily or weekly.
- Super Loops expose pause/resume/stop; desktop/mobile can steer and cancel sessions and answer prompts.
- Nested workstreams and session kanban show parent/child activity and rollups; no high-load fixture was run.
- Direct sessions can grow into orchestrated workstreams with retained caller links.
- Local desktop execution plus user-controlled headless Node/Cloudflare sandbox execution are documented, with remote…
- Parallel coding sessions and loops can use separate Git worktrees.
- Desktop auto-continue and persisted session history are documented.
- Session states, workstream rollups, logs and context usage are visible; no operational accuracy benchmark was run.
- Humans collaboratively edit shared documents/trackers with presence and discussion.
- Team invite/join and owner/admin member/invitation controls are documented; SSO assurance remains unestablished.
- Local interactive permission prompts, plan approvals and per-project agent permissions are available.
- Tracker states, document revisions and commit/session links retain activity and decisions.
- Rendered document changes and code/PR diffs support accept/reject and review navigation.
- Documents, code diffs, mockups, browser captures and tests can remain linked to sessions/trackers.
- Built-in Git worktree, commits and GitHub PR review/merge are documented.
- In-app PR approve and merge uses gh authentication.
- Commit history links produced commits to AI sessions and retained tracker references.
- Application repository is MIT licensed; separate collaboration server license/parity is not established.
- Core desktop and headless runner can run on user infrastructure; hosted team sync is a separate service/project.
- PostHog anonymous analytics can be disabled.
- The Nimbalyst app and personal wiki are free.
- v0.82.7, published November 4, is marked a GitHub pre-release; the latest release not marked pre-release is v0.81.7,…
- Free application download with MIT source is available on desktop.
Sources
Open the Nimbalyst page
open-sweinitial research34 findings · 11 sources
Findings today
- Experimental desktop installation supports Apple Silicon macOS Ventura or later.
- Robin runs its own Deep Agents loop on Managed Deep Agents; the Claude Code connection is a read-only mirror.
- Deployments use model-provider API keys or LangSmith Gateway credentials.
- Setup documents backend deployment, GitHub App, Slack App, allowlists and model credentials.
- robin/oswe are product CLI entry points.
- Product-managed instance, workspace and personal MCP connections support headers or OAuth; collaborative threads…
- Robin Desktop uses Electron and downloads a backend-selected UI archive.
- Work starts in the dashboard, GitHub conversations, Slack or optional Linear delegation.
- Shared repository instructions, organization skills and private user instructions have documented scopes.
- Independent cloud threads and separate local worktrees can run in parallel.
- Local desktop refuses starting another active agent in the same tree or switching its branch underneath it.
- The runtime configures general-purpose subagents for independent work.
- Recurring automation can start system-owned coding threads.
- Slack Stop interrupts ordinary thread work, while other participants queued work may continue.
- This Mac bridges a selected local folder to the backend-managed agent loop.
- Local desktop can create a dedicated Git worktree per thread.
- Local threads keep their transcript on the backend and can reopen the bridge on follow-up.
- Admins can capture workspace toolchain snapshots for later sandboxes.
- LangSmith is the default trace provider.
- Workspace and Slack collaborative threads can be prompted by participants without using personal MCP credentials.
- Configured admins administer all workspaces; signed-in users can use every workspace.
- Workflow-file push approvals cover detected pushes rather than every shell/API write.
- Opted-in dashboard mutations and marked agent tools persist actor-attributed activity; unmarked operations are excluded.
- Opt-in Slack cards show a PR summary, size, branches and reviewers with GitHub review links.
- Desktop Changes compares against the initial Git snapshot and links the discovered PR.
- Human-review requests are refused for drafts, conflicts or failing required checks; this is a review-request gate.
- Robin opens and updates PRs, runs reviews and responds to CI feedback.
- Configured standard Slack human review can merge once reviewers approve, subject to its policy and feature flag.
- Review requests and desktop Changes link the relevant PR.
- Optional merge queue observes GitHub candidates; native observation does not execute merges or tests.
- Robin code is MIT licensed; vendor/ material is outside this repository grant.
- The core backend and web UI can be deployed using the documented Docker image.
- The documented optional PostHog and Segment integrations send nothing without keys.
- The project warns of breaking changes and calls desktop experimental; web UI is recommended.
Sources
Open the open-swe page
OpenChamberinitial researchresearch correction41 findings · 45 sources
Findings today
- OpenChamber has macOS, Windows and Linux desktop clients, a web/PWA client, VS Code extension and a CLI/server.
- The core workspace uses OpenCode to run coding agents.
- The Claude Code integration reuses Claude Pro or Max through the local Claude Code CLI sign-in without API keys.
- Desktop bundles OpenCode; CLI/Web requires Node.js 24.14+ and an installed OpenCode.
- Product CLI commands expose status, connect QR links, tunnels, startup control, logs, stop and update.
- Users configure local command or remote URL MCP servers with headers/environment settings, scoped personally or per…
- Git host issues and PRs become context for a new worktree/session; the message is not sent automatically.
- Plans are editable Markdown files retained at project level, imported from text/Markdown and optionally moved into…
- Agents can link todos to a plan and add dependencies on other todos in that plan.
- Project notes can be pinned into a session.
- Changes Walkthrough groups code-related edits, reads code/docs/tests and anchors its explanation to exact diff…
- The native project todo checklist persists across sessions, supports checked completion and ordering, and can…
- The managed-server project knowledge tool lets agents read and edit the same notes, plans and todos humans see, link…
- Multi-run starts several model sessions from one prompt, optionally each in its own worktree/branch.
- The native control tool creates sessions, sends prompts or follow-ups and can create worktree sessions.
- An agent can create a child session with a selected model, list status, read the latest assistant response and send…
- Native scheduled tasks send prompts in new sessions once, daily, weekly or by cron, optionally as goals.
- Stop and Pause abort the current turn and pause a goal; Resume re-arms it.
- The In work sidebar groups ongoing sessions and worktrees, shows status and news, and permits marking work done or…
- Natural-language requests can create another model session, an isolated worktree or a schedule, and inspect results…
- OpenChamber can run as desktop app or CLI/server on a workstation or the user's own server.
- A worktree session has its own checkout and branch to isolate concurrent edits; optional multi-run isolation creates…
- Archived worktrees preserve sessions, uncommitted changes and branch data for restore, but ignored files such as…
- Project setup commands run after worktree creation and on restore.
- Goals have optional token budgets, an automatic-continuation cap and stop on turn errors.
- The UI shows working, waiting, finished and failed sessions with approvals, schedules, provider limits, token use…
- Usage shows provider windows, model breakdown and pace; the agent usage tool can query those provider numbers and…
- OpenCode permission rules allow, deny or ask.
- Agent-created notes are labeled and stale whole-document replacements can be refused against the prior text.
- A guided walkthrough groups related edits into ordered stops with code anchors, shows outdated/uncovered areas and…
- Preview places the running app beside chat and can send element screenshots, styles and errors as agent context.
- A separate model auditor compares the latest turn to the goal, continuing until done, blocked or budget/turn limits.
- The Git panel can create/update PRs, mark drafts ready and merge, with equivalent merge requests or Gerrit changes…
- Users merge from the PR panel; Gerrit Submit waits for the change submit requirements.
- Walkthrough explanations are bound to exact code content.
- The OpenChamber application source is MIT licensed at the pinned revision.
- The full workspace/server can run on a user workstation or own server; optional Private Relay and tunnels provide…
- A Send anonymous usage reports switch in Settings turns off usage reporting, and Enterprise mode forces it off.
- The source application is available under a free-of-charge MIT grant.
- OpenChamber v2.7.0 was published November 8.
- The core application is free MIT software with user-hosted execution.
Sources
Open the OpenChamber page
ralph-claude-codeinitial research30 findings · 9 sources
Findings today
- Ralph for Claude Code supplies terminal CLI commands and a tmux status monitor on documented macOS/Linux…
- The shipped loop runs Claude Code; multi-provider abstraction is still listed as in progress.
- Docker execution accepts ANTHROPIC_API_KEY or a scoped copy of existing Claude CLI credentials.
- Install once, enable or import a project, review .ralph files and run ralph --monitor.
- The product CLI inspects state, manages queues, imports work and controls loop settings.
- PRD imports populate .ralph prompt, specs and task checklist files.
- A local PRD file can be added to the persistent queue without a GitHub issue.
- fix_plan.md holds prioritized checkbox tasks while specs holds detailed requirements.
- depends on, blocked by and requires references inside issue bodies become queue dependencies.
- PROMPT.md, AGENT.md and specs/ supply project requirements and build guidance.
- Local PRD queue entries can exist before explicit processing starts.
- Documented CLI reads persistent queue status and next ready item.
- Duplicate queue IDs are skipped; ready items run in priority/FIFO order.
- Ralph checks completion inside a manually started loop; the product has no schedule or external-event intake that…
- Ctrl+C interrupts the monitor; loop termination and session reset are separately documented.
- The loop and default Claude execution run on the current host.
- Ralph offers optional Docker execution and an E2B Linux-VM sandbox.
- E2B expiry can recreate the sandbox and upload project files again.
- E2B bootstraps Claude CLI on the default template or accepts a prebuilt custom template.
- Claude invocation timeouts are configurable from 1 to 120 minutes.
- The monitor shows local logs, loop count, API-call usage and queue totals from JSON state.
- E2B sandbox cost uses an estimated hourly rate, not a reconciled provider invoice.
- The queue is sequential on one branch, not a documented parallel fleet scheduler.
- ALLOWED_TOOLS comes from repository-controlled configuration and is not a boundary against that repository.
- Local queue state, logs and the last 50 session transitions persist on disk.
- Queue processing commits one commit per issue on the current branch.
- --backup creates Git backup branches and --rollback restores a selected backup.
- The distributed shell application source is MIT licensed.
- The loop, queue and monitor run locally from the public source distribution.
- README reports active development at v0.11.5 with newer pinned-source changes described as unreleased.
Sources
Open the ralph-claude-code page
ralph-orchestratorinitial researchresearch correctionfreshness recheck27 findings · 9 sources
Findings today
- Ralph supplies a native CLI/TUI on macOS, Linux and Windows plus an alpha browser dashboard and Telegram integration.
- Eleven named CLI backends and custom commands are documented, with different backends selectable per hat.
- Backend setup supports native Claude login and provider API credentials.
- Install Ralph and a supported agent CLI, initialize a backend, then plan or run.
- Ralph provides its CLI and a stdio MCP server scoped to one workspace.
- Runtime tasks can be added, listed and inspected through product commands.
- Interactive planning writes requirements, design and implementation-plan Markdown under .ralph/specs.
- Persistent Markdown memory is shared across parallel loops; tools can search, add and delete entries.
- Ralph owns persistent runtime task records with ready, close and fail operations; a full issue backlog workflow is…
- Runtime task commands are exposed to agents as tools.
- Planning creates persistent requirements and design files; approval and drift gates remain unestablished.
- Additional concurrent loops run in separate Git worktrees.
- The shipped Hatless Ralph coordinator delegates work through named hat events, receives their event results and can…
- Loops can be stopped/resumed; Telegram questions block until answered and proactive messages steer a run.
- Five built-in workflows and direct prompts are documented.
- Ralph normally invokes installed agent CLIs on the current host.
- Ralph separates parallel loop checkouts with Git worktrees and runs native host CLI processes; it provides no…
- Existing loop state can be continued and suspended loops resumed; no crash recovery experiment was run.
- Current configuration bounds iterations and total runtime, with adapter inactivity timeouts and per-hat activation caps.
- Loop logs and event history expose operational state and JSONL export.
- Planning documents and branch diffs can be retained, and publish-review writes a Markdown handoff.
- Backpressure workflows request test, lint and typecheck evidence; instruction examples do not prove that an external…
- Ralph integrates worktree diff, branch publish/rebase and merge operations; publish-review writes a Markdown handoff…
- Optional auto-merge uses merger, resolver and tester hats.
- Ralph application source is MIT licensed.
- The core Ralph CLI and local web stack can run on user-controlled hardware.
- The latest listed release is 2.10.1, published June 23; current README labels the browser dashboard alpha.
Sources
Open the ralph-orchestrator page
Ralph TUIinitial researchfreshness recheck31 findings · 6 sources
Findings today
- Ralph TUI is a native terminal CLI/TUI with documented macOS, Linux and Windows installation and remote terminal…
- Release v0.12.0 names Claude Code, OpenCode, Factory Droid, Cursor CLI, Gemini CLI, Codex and Kiro CLI, plus a…
- Install with Bun, run setup, create a PRD and start from prd.json or a Beads epic.
- Product-owned CLI supports run, resume, status, logs, configuration, tracker conversion and remote management.
- Agent/tracker plugins and installable PRD/task skills extend the orchestrator.
- Terminal interface uses OpenTUI/React and Bun; no Electron desktop client was established.
- Intake supports local prd.json and external Beads, plus conversion into tracker formats.
- Interactive AI-assisted PRD creation and progress files retain plan/context between iterations.
- One graph spans selected epics and blocks dependent work on incomplete external dependencies.
- Product progress file retains cross-iteration context and custom skills/templates can be selected.
- Optional parallel worker pool executes independent tasks simultaneously in worktrees.
- Scheduler selects highest-priority ready tasks and dispatches them from a dependency graph.
- Ralph TUI has a deterministic parallel worker engine and provider-subagent tracing, without a product-supplied agent…
- Keyboard and remote actions pause, resume, interrupt or cancel execution, with persisted logs.
- Multi-epic sessions show source scopes and shared workers in a single scheduler UI; no fleet usability fixture was run.
- Agent/model selection, custom templates and task-tracker plugins configure execution.
- Run engine on the local machine or a user-controlled VPS/CI host and control it from another TUI.
- Ralph TUI uses Git worktrees for parallel file separation and optional bubblewrap/macOS sandbox-exec process policies.
- Sessions can be resumed and remote connections use bounded reconnect retries.
- Iteration limit bounds loop count; preflight timeout bounds setup checks, not all runtime/tool effects.
- Dashboard displays agent/model, tracker, branch, sandbox and remote status; log files retain iteration output.
- Worker pool and dependency grouping bound concurrency; OS sandbox explicitly does not enforce CPU/RAM/process limits.
- Remote tokens and opt-in sandbox paths/network constrain access; no enterprise approval workflow established.
- Remote actions are logged to a local audit.log; immutable retention is unestablished.
- PRD/progress and per-iteration logs persist on disk.
- Ralph TUI integrates worktree merging, conflict handling and optional commits, then prints a manual gh pr create…
- Orchestrator merge queue brings completed worker branches back sequentially.
- Backup tags are created before each merge for rollback safety.
- Ralph TUI application source uses MIT license.
- Core CLI/TUI engine and authenticated listener run on user-controlled machines.
- Latest listed release v0.12.0 was published May 13, 2026; current default-branch documentation may include later…
Sources
Open the Ralph TUI page
ralphexinitial researchresearch correctionfreshness recheck30 findings · 6 sources
Findings today
- Ralphex is a primary native CLI on Linux/macOS and best-effort Windows source builds, with an optional responsive…
- Claude Code and Codex have first-class executors, several named agents have shipped wrappers, and the first-party…
- Provider authentication is inherited from agent CLIs; Claude headless subscription use has a separate billing warning.
- Run from a Git repository with a plan and installed coding agent.
- CLI configures plan creation, execution, review, worktrees and dashboard.
- Custom prompts, review agents and provider configuration extend the workflow.
- Markdown plans are authoritative local execution input.
- Plans retain validation commands, tasks and checkbox progress; paused work can be amended.
- Plan tasks execute sequentially with configurable multi-phase reviews.
- Repository guidance and review prompts can be configured per project.
- Parallel review agents and independent worktree plan runs are documented.
- Ralphex owns deterministic task/review phases and ships reviewer prompts, but delegation uses Claude Task or Codex…
- Unix signal controls pause/abort execution; notifications are optional.
- Dashboard watches multiple plan progress files without a verified epic-management fixture.
- Modes and per-phase model settings are configurable; workflow changes are mostly configuration-driven.
- Ralphex runs installed agent processes on the operator machine or Docker host.
- Ralphex offers Git worktrees and a Docker wrapper.
- Plan progress and interrupted worktrees survive, but context is intentionally fresh per task.
- Iteration, review-patience, timeout and rate-limit controls are documented.
- Timestamped logs and dashboard expose progress; this is not a tested state-consistency result.
- Default Claude invocation bypasses permission prompts; safer host isolation is optional.
- Plan, commits, progress logs and reviewer output are retained.
- Default pipeline requests validation and multiple review phases; tasks-only and optional stages limit enforcement.
- Ralphex creates branches/worktrees and commits, and its optional finalize prompt rebases/organizes commits.
- Finalize is optional customizable agent automation, not a tested delivery authority boundary.
- Completed plans and commits persist; success has a documented archive caveat.
- Repository source is MIT licensed.
- Core standalone CLI requires no product cloud service.
- Public release and issues exist, with explicit best-effort Windows support.
- Open-source binary and source allow evaluation with user-provided agents/compute.
Sources
Open the ralphex page
Replitinitial researchresearch correction41 findings · 44 sources
Findings today
- Replit offers browser and phone-browser access, native iOS/Android clients, and native desktop apps on Mac, Windows…
- The documented development experience runs Replit Agent with selectable modes/models.
- Replit plan credits pay for Agent.
- Start a project by describing an idea in plain language, or bring it in from GitHub, Figma or uploaded files, in the…
- A main-thread prompt can create titled tasks with descriptions and plans, or users can start a planning thread.
- Agent proposes detailed plans and definitions of done; users inspect View plan, accept tasks or request a revised…
- Agent identifies prerequisite tasks and queues dependent work until the prerequisite is applied.
- Each background task inherits the project context and receives its own Agent and project copy.
- Projects can begin from GitHub, Figma or uploaded files; this establishes input channels, not verified architecture…
- Native tasks have Drafts, Active, Ready and Done states, a board, separate threads and explicit acceptance.
- Agent can split a request into tasks and put them on the human-visible task board.
- Core supports one active background task, Pro up to ten and Enterprise up to sixty-four.
- Agent schedules independent tasks when capacity exists and holds dependent tasks until prerequisites are applied.
- Replit Agent decomposes a request into background Agent tasks with separate threads and returns work logs, tests and…
- Paid-plan chat Routines repeat hourly, daily or weekly without a new request.
- Cancel stops building and discards its work.
- A project board displays Drafts, Active, Ready and Done tasks alongside task conversations.
- Users start with ordinary chat, refine plans and let Agent create task lists before approving work.
- Replit provides managed cloud development, and the current desktop page also offers local Agent code/build execution…
- Replit repository development sandboxes explicitly use microVMs.
- Files, history and running app are preserved in the cloud for a later visit.
- Agent sets up the project and infrastructure.
- Core and Pro Routines have per-run budgets checked after each step; an in-progress step can complete beyond the budget.
- Finished tasks expose a work log, test results and preview; active threads and board cards show task status.
- Agent billing is effort-based and uses plan credits shared with cloud services.
- Plan-based task limits cause additional background tasks to queue until slots are available.
- Invited teammates work on one project through separate Agent threads and shared task boards.
- Personal workspaces have an admin and project guests; team workspaces add members with access to all workspace projects.
- Humans approve plans before background execution and apply or dismiss finished changes before they affect the main…
- Replit displays code diffs in its Changes pane, alongside local changes or linked PRs, in addition to task logs,…
- Task review retains a work log, test results and a live preview in the task flow.
- With App Testing enabled, Agent periodically tests apps it builds in an actual browser and fixes issues it finds.
- The repository contribution workflow creates branches, commits/pushes code and opens draft GitHub PRs or GitLab MRs,…
- Users publish to a live URL from the Project Editor.
- Agent creates checkpoints that users can roll back to when a change is wrong.
- Replit imports/exports project artifacts and offers Enterprise deployment options, but even Full BYOC explicitly…
- Pro users can opt out of Model Improvement in Personalization settings, and Replit states it does not train on…
- Replit supplies recurring Core/Pro plans and metered Agent/cloud credits.
- The November 6 changelog documents Agent-assisted support investigation and Pro ticket submission.
- Starter offers recurring daily Agent credits capped monthly and monthly cloud credits, with one published app whose…
- Agent uses effort-based paid pricing with plan credits; actual charges depend on request complexity and cloud services.
Sources
Open the Replit page
scioninitial researchresearch correctionfreshness recheck32 findings · 12 sources
Findings today
- Scion provides native macOS/Linux CLI and Windows Hub-client CLI distributions, Workstation/hosted browser UI and…
- Named harnesses include Claude Code, Codex, OpenCode and Antigravity; credential guide explicitly includes Gemini CLI.
- Workstation injects user native harness login files or API keys; hosted broker credentials must be explicit.
- Onboarding detects container runtime, identity and images; init links project to Hub.
- Product CLI and Hub JSON APIs expose agent/project lifecycle and messaging with scoped authentication.
- Templates, versioned Skill Bank, lifecycle hooks, MCP configuration and A2A bridge extend integrations.
- Hub owns projects and agent tasks/summaries, but companion issue trackers remain separate tools.
- Templates, versioned skills and shared directories provide team context; specialized memory systems are companions.
- Agents run concurrently on local or remote infrastructure with per-broker admission limits.
- Agents can delegate through CLI and operate under recorded parent/creator authority.
- Scion provides agent creation, readable descendants, messaging and completion/attention visibility through product…
- Hub-owned cron schedules start recurring agents/messages with timezone-aware timing and recorded authority.
- Users message agents and inspect logs, stop/resume sessions and receive attention notifications.
- Web/CLI list and inspect agent fleet with phase, lineage and status.
- Users define roles and collaboration in instructions/skills rather than a fixed orchestration graph.
- Scion supports local Workstation or user-operated durable hosted Hub/brokers on VM, Kubernetes or Cloud Run.
- Scion has container and Git worktree/clone separation.
- Agents can stop/resume; hosted durability differs between single-node SQLite and replicated Postgres.
- Runtime onboarding, container images and credential provisioning prepare execution environments.
- Hub refuses starts on unresolved credentials and enforces per-broker running-agent admission; provider usage/spend…
- Hub exposes phase/activity, logs, session metrics and system health.
- Hub limits running agents per broker, with reservations reconciled at startup/hourly.
- Feature-flagged native chat allows humans and agents in shared project threads, with a default agent and…
- Hosted users have admin/member/viewer roles and project role bindings, with revocation/suspension and…
- Scoped agent tokens, role bindings and narrowing access constraints govern Hub operations; this is not proof of…
- Access-boundary mutations retain transaction-bound audit history; deletion removes boundary history, so immutable…
- Experimental artifact review supports retained marks and versioned review outputs.
- Experimental default-off artifact service stores versioned files/folders on Hub and exposes CLI/web access and review.
- Scion core application/platform source uses Apache-2.0.
- The core Hub, clients and brokers run on user infrastructure in Workstation or hosted mode.
- Scion supports explicit hosted credentials and optional workstation credential injection.
- Current nightly November 8 and preview releases document pre-1.0 actively evolving platform.
Sources
Open the scion page
Supacodeinitial researchfreshness recheck22 findings · 5 sources
Findings today
- Supacode ships a native macOS 26+ application and its local CLI.
- Supacode runs coding CLIs in real terminal panes and ships presence/skill integrations including Claude Code, Codex,…
- Local plain folders and Git repositories are supported; source builds need macOS/Xcode/mise and submodules.
- Product CLI and deeplinks address repos, worktrees, tabs and surfaces; session IDs are exported.
- Native macOS client uses libghostty with The Composable Architecture and zmx.
- Several coding agents run side by side, with one worktree/terminal per task.
- Agent presence badges, attention notifications and worktree hoisting expose need for input; terminal input provides…
- Worktrees can be created by UI, CLI or deeplink, and named repo/global scripts run in dedicated tabs.
- Local execution and beta SSH-host execution are documented.
- Supacode separates tasks with Git worktrees and terminal sessions.
- zmx sessions are independent of app lifecycle, retain scrollback and reconnect after relaunch.
- Per-repo setup/archive scripts and named custom scripts are supported; environment repair unestablished.
- Presence and live branch/file/PR state help track work.
- Supacode retains terminal output and Git/PR state.
- Worktree management and GitHub PR actions show PR state, checks and merge readiness; site explicitly says open PRs.
- GitHub integration exposes PR merge actions and configurable merge strategy.
- Supacode application source uses FSL-1.1-ALv2.
- Core native application and local/remote zmx run on user hardware; commercial activation is required beyond trial…
- One-time USD 99 Basic, USD 129 Standard or USD 79 per Team seat, minimum three.
- Latest listed release v0.11.7 was published November 7; security support targets latest release.
- Five local worktree creations are trial access; existing worktrees/terminals stay usable and SSH worktree creation…
- Public lifetime-license prices, seat counts, trial quota and update renewals are stated.
Sources
Open the Supacode page
Traycerinitial researchresearch correctionfreshness recheck38 findings · 17 sources
Findings today
- Traycer Desktop ships macOS/Windows/Linux clients, native iOS/Android companions and a product CLI.
- Traycer documents multiple coding agents and model switching with shared context.
- Existing coding-agent subscriptions are supported; optional Traycer inference has separate billing.
- Install Desktop, sign in, prepare the Host and select agents.
- Traycer has task, agent, worktree and host commands with scoped agent access.
- Traycer manages MCP configuration per provider; availability differs by provider.
- Tasks are local work containers associated with one or more folders.
- Built-in skills create durable brief, technical-plan and ticket artifacts.
- Ticket breakdown and execute describe dependencies and batched work through skills.
- Tasks retain planning artifacts and support built-in planning and review skills.
- Traycer retains Tasks independently of their multiple agents and provides organization in History.
- Requirements revision is a documented skill, but persistent product-wide conformance is untested.
- Parallel agent work is documented, with visible status in Agent Office.
- Agents can claim named roles when the optional Agent roles setting is enabled.
- Traycer-owned agent commands support cross-harness delegation, transcripts and replies.
- Traycer documents an external host-cron recipe that creates nightly Tasks with idempotency keys.
- Agent Office supplies lineage, messages and state views; the 12-task fixture was not run.
- Optional skills and per-agent configuration support different workflow sizes.
- Agent execution can use the local host or a user-provisioned remote host.
- Separate Git worktrees are documented; this source does not establish a security boundary.
- Transcript durability is documented separately from availability.
- Worktree setup/teardown scripts support dependency setup with visible failure cards.
- Tasks, artifacts and shared chats are visible to collaborators, but shared prompting is explicitly restricted.
- Task Owner/Editor and team Admin/Member roles are documented; SSO/SAML is unavailable.
- Agent creation exposes permission modes with full_access as the default.
- Review skills retain findings and a walkthrough, while artifacts support discussion.
- Specs, tickets, reviews and walkthroughs are durable Task artifacts.
- Execution/review skills propose verification; the source explicitly denies automatic enforcement.
- Traycer integrates worktrees and PR state; feedback can be sent to the associated agent.
- PR rows link Tasks/chats to commits and delivery state; rollback was not tested.
- PR stacks and review summaries support browsing several changes, without a tested shipping queue.
- Repository source is MIT licensed; hosted parity is not established.
- Hosts run on user machines, but account, relay and cloud-sync services remain vendor dependencies.
- Traycer discloses product analytics and separate provider requests.
- Free product access is separate from inference and paid team-wide sharing.
- Desktop release and first-party support channels are public; SLA was not established.
- Recurring Free includes own agents, sync, remote hosts and mobile; whole-team sharing requires Sync.
- Public plans separate product access and optional inference credits.
Sources
Open the Traycer page
Archived event links
Something wrong or missing?