November 2026 status archive, page 10 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
2 ADEs
Zencoderresearch correction2 findings · 3 sources
18 ADEs
Agent Deckinitial researchresearch correctionfreshness recheck33 findings · 4 sources
Findings today
- CLI/TUI ships for macOS and Linux; Windows access uses WSL.
- Many terminal coding agents are supported with different integration levels.
- Agent authentication is reused for local/container sessions, with a macOS Claude exception.
- Install binary and add a directory with an installed tool; remote setup can install deck on SSH host.
- CLI exposes session creation, send, stop, fork, parent relationships, MCP, skills, costs and remotes.
- Product manages MCP and skills per supported coding tool.
- Sessions carry semantic goal/state/ticket annotations but external ticket references do not establish native issues.
- Session parent linkage represents delegation, with a documented remote-parent boundary.
- Conductor identity, shared policy and learnings persist in files.
- Fleet board cards and work states are session annotations.
- Independent tmux-backed sessions run concurrently and show process/work state separately.
- Conductors dispatch and monitor sessions; native issue claims were not established.
- Persistent conductor sessions orchestrate workers and receive status nudges.
- Heartbeat/watchers and optional chat-service bridges trigger session work.
- Conductor auto-responds or escalates; users can send and stop sessions.
- TUI and web Fleet/Command Center expose grouped session state and attention.
- Users can work directly in sessions or enable conductors and configure tool/model choices.
- Sessions run locally, in Docker or on user-owned SSH/cloud hosts.
- Git worktrees and Docker containers separate execution scope.
- Session resume/fork support is tool-specific; remote failures are visible and retryable.
- Worktree hooks initialize task environments; Docker setup may require separate authentication.
- Cost budgets and run token/turn/time defaults are configurable, with explicit untested enforcement limits.
- Status lights and semantic annotations are separate; roadmap lists detection gaps.
- Session costs and tokens are collected from provider transcripts/output, with limited provider verification.
- Web mode has token and read-only controls; shared token is not organization RBAC.
- Conductor task-log and semantic decisions persist, without an audit integrity guarantee.
- Transcripts, forks and archives retain session work.
- Product manages worktrees and branches.
- Agent Deck repository is MIT licensed.
- Core TUI, web server and conductor run on user-controlled machines.
- Usage telemetry is opt-in and has CLI preview/disable controls.
- Public releases and community support are available.
- MIT source/direct binary installs support local evaluation with own tool accounts.
Sources
Agent of Empiresinitial researchresearch correction25 findings · 13 sources
Findings today
- aoe serve starts the released browser dashboard.
- Docs list Claude Code, OpenCode, Codex CLI, Gemini CLI, Cursor CLI, Pi and other agent CLIs as supported and…
- Agent credentials and native stores remain harness-specific.
- Install AoE and tmux on Linux/macOS, then add an installed agent or start the dashboard.
- CLI includes add, list/ps, send, wait, fork and session lifecycle operations.
- Configured MCP server definitions are forwarded to structured-view agents when a session starts; terminal sessions…
- Users create local agent sessions through CLI, TUI or dashboard.
- Profiles, repository configuration/hooks and custom agent commands configure session behavior.
- Agents run in parallel across separate branches/worktrees.
- Session lifecycle and global killall controls stop workers; killall is explicitly destructive and unprompted.
- Structured view supports explicit model and reasoning effort choices subject to cached provider options.
- AoE runs agents on the operator local or SSH host and optionally the configured container runtime.
- Optional worktrees and Docker/Podman/Apple Container sandboxes separate sessions.
- The TUI can reattach persistent tmux sessions after reconnect.
- Sandbox seeding uses private agent stores and a shared Claude credential file with documented refresh behavior.
- TUI/browser status detection and CLI ps provide a runtime session overview.
- Optional per-container CPU and memory limits are configurable; no global compute-aware queue is established.
- An MCP client has the same access as the keyboard user; AoE adds no per-tool permissions.
- Diff line comments can be edited and sent as a follow-up prompt with the captured snippet.
- The browser/TUI diff viewer displays changes against selectable comparison bases.
- AoE manages session worktrees and branches, without a built-in PR creation/management workflow in this scoped core.
- AoE uses MIT with a separately documented Apache-2.0-derived hyperlink file.
- The full daemon, session manager and optional web dashboard run on user-controlled Linux/macOS hosts.
- Anonymous telemetry is off by default and requires opt-in.
- README links public issues, documentation and community channels.
Sources
- Agent of Empires README.md
- Agent of Empires docs/cli/reference.md
- Agent of Empires docs/guides/web-dashboard.md
- Agent of Empires docs/api.md
- Agent of Empires docs/guides/mcp-control.md
- Agent of Empires docs/guides/mcp-servers.md
- Agent of Empires docs/guides/sandbox.md
- Agent of Empires docs/guides/worktrees.md
- Agent of Empires docs/guides/session-resume.md
- Agent of Empires docs/guides/diff-view.md
- Agent of Empires docs/telemetry.md
- Agent of Empires LICENSE
- Agent of Empires v1.22.1
Agent Teamsinitial researchresearch correctionfreshness recheck35 findings · 5 sources
Findings today
- Desktop supports macOS, Windows and Linux.
- Claude Code, Codex and OpenCode are managed runtimes.
- Use existing provider subscriptions or API keys, or start with a free unauthenticated model.
- Download desktop app, choose project, define team roles/provisioning prompt and launch.
- Built-in MCP server and agent plugins support tool integration; universal secure plugin system remains roadmap work.
- Desktop app uses Electron 40, React 19, TypeScript, Tailwind and Zustand.
- Humans create tasks on the board or send attachments to the lead, which creates described tasks.
- Task descriptions, comments and attachments retain assignment context.
- Organizations can nest teams and show dependencies/delegation.
- Instructions and task context survive compaction through app recovery mechanisms.
- App owns shared tasks with descriptions, comments, attachments and durable board status/history.
- Agents create/manage shared tasks and leave comments visible to humans.
- Multiple teams across projects and their agents can work in parallel.
- Agents plan, delegate and create shared tasks through the app's orchestration layer, and the team lead turns a human…
- App-owned lead/teammate orchestration exposes cross-team direct/group messages, task comments and review.
- App-owned scheduler dispatches due persisted schedules through a runtime process executor.
- Direct messages course-correct agents, diagnostics/restart address stuck teammates, and board badges flag human input.
- Organization map shows nested teams, agent status, tasks, dependencies and cross-team messages; no high-load fixture…
- Teams may begin in solo mode, expand later, and use provisioning prompts to set roles and branch rules.
- Teams execute as local processes on the desktop host.
- Selected teammates can get worktrees at launch.
- Task context persists; app has compaction recovery and guarded nudges after known rate-limit cooldowns.
- Monthly token/estimated-cost budgets offer threshold alerts; supported scheduled Claude runs can receive a…
- Per-task logs show tool actions and messages; analytics and live process/resource sections expose activity.
- Token analytics include input/output/cache/reasoning and estimated cost across tasks, agents and runs.
- Users can require manual approval for supported tool actions, with readable requests and notifications.
- Task timeline records state changes, participants and workflow actions.
- Task diff review supports accept, reject and comments on individual hunks.
- Tasks retain attachments and matched runtime logs/changes, with document and code preview/editing.
- Main application public repository uses AGPL-3.0.
- Core desktop app and local orchestration run on user hardware; app does not have a cloud backend for storing project…
- App says it does not upload/sync project code to Agent Teams servers.
- Application is free.
- Latest listed v2.18.6 was published November 5; current release requires macOS 13+, with older 2.17.1 for macOS 12.
- Free app can begin with a free model requiring no signup, key or card; other providers retain their costs.
Sources
ai-devkitinitial researchfreshness recheck29 findings · 11 sources
Findings today
- AI DevKit has native macOS/Linux CLI/TUI operation; the managed console is unsupported on Windows.
- Managed start supports named coding harnesses and supervision across providers; configuration integration alone is…
- Underlying agent logins/provider credentials supply access.
- Machine setup installs skills/hooks/memory wiring; init writes project config.
- CLI exposes managed agent control, memory and daemon operations; memory MCP server provides tool access.
- Unified config reconciles per-agent instructions/skills/MCP setup; compatible agents receive memory-server wiring.
- Lifecycle skill writes requirements/design/planning/implementation/testing files under docs/ai and requires staged…
- Local SQLite memory supports titled/tagged/scope records through MCP and CLI, retrieved when relevant.
- Memory retrieval supplies prior project knowledge on demand.
- Lifecycle skill persists requirements/design/plans in repo Markdown; no server-enforced spec-drift gate established.
- Several agents can run in separate managed sessions and be supervised together.
- Lead uses named workers and start-then-assign handshake; same-folder session detection collisions are documented.
- AI DevKit orchestration skills equip a lead to start, message, inspect and wait for workers across harnesses.
- Console messages or Telegram/Slack bridge can steer running agents; managed kill requires confirmation.
- Composable lifecycle/verification/orchestration skills guide workflows without prescribing one fixed agent.
- The local CLI/daemon runs agents on its own macOS/Linux host, including a user-controlled always-on Linux server…
- Worktree setup is manual guidance: users create directories and pass --cwd.
- Managed durable mode is limited to Claude/Codex/Pi; daemon persistence can be installed on Linux.
- send --wait supports timeout and readiness handshake; quotas are readings, not enforced spend limits.
- Console shows last-observed running/waiting/idle/unknown status and transcripts; status readiness is explicitly a hint.
- Capacity reports provider quota windows and reset times.
- Slack proof-of-concept bridge permits multiple workspace members to send text toward one agent; Telegram allows only…
- Underlying harness permissions/trust must be configured; tested presets are Linux 0.70.1 guidance with broad access.
- Phase Markdown and historical transcripts remain available locally; output can be piped into agent send.
- verify/TDD/review skills instruct fresh verification before completion; no independent runtime-enforced gate was…
- AI DevKit application source is Apache-2.0 licensed.
- Core CLI/daemon, memory and console run on user-controlled machine; no hosted service required.
- README explicitly says no telemetry and local memory.
- Latest listed CLI release 0.72.0 was published November 3; console remains experimental macOS/Linux and unsupported…
Sources
- README.md
- web/content/docs/13-agent-console.md
- web/content/docs/17-multi-agent-orchestration.md
- web/content/docs/6-memory.md
- web/content/docs/15-capacity.md
- web/content/docs/16-runtime.md
- web/content/docs/19-permissions.md
- web/content/docs/12-channel.md
- web/content/roadmap/10-agent-management.md
- LICENSE
- 0.72.0
Aletheinitial researchfreshness recheck37 findings · 7 sources
Findings today
- Desktop supports Windows/macOS/Linux; optional LAN browser remote control has separate input opt-ins.
- Multiple agent CLIs run in independent PTYs, with cross-agent context handoff.
- Agent credentials remain in underlying CLIs; provider quota reads use native signed-in accounts.
- Desktop installers and agent CLI launcher detection support setup; first-task timing was not measured.
- Product manages coding-agent MCP and skills plus a permission-declared plugin system.
- Desktop is Tauri/Rust with React and xterm.js, not a fully native widget UI.
- Projects organize local work and sessions; optional todos and orchestration are separate capabilities.
- Official orchestration skill writes bounded briefs and result artifacts.
- Playbook canvas supports worker/check/condition graphs; task-DAG roadmap is a separate effort.
- Local session/history and MCP/skill state persist; handoff uses a bounded context packet.
- Graphify exposes a project code graph through MCP.
- Bundled Todo List persists per-project tasks separately from agent PTYs.
- Independent agent PTYs and opt-in orchestrator workers run in parallel.
- Opt-in MCP coordinator delegates worker briefs; arbitrary DAG dispatch is not inferred.
- Opt-in MCP-delegate orchestrator supports Claude planner to Codex workers, with an optional standalone official skill.
- Worker steering/approvals, PTY input and group suspension support intervention.
- Project panes and opt-in worker board expose active sessions, without a tested 12-state fleet fixture.
- Direct PTYs coexist with optional playbook delegation; not all task-DAG roadmap work is shipped.
- Agents execute on the local desktop host.
- Git worktrees and provider-scoped approval controls are implemented.
- Workspace and supported agent sessions survive layout changes/restart; context transfer has limits.
- Resource, activity, worker cost/token and quota displays are documented.
- Orchestration reports use CLI-reported worker spend; no provider bill reconciliation was performed.
- Remote Control defaults read-only; agent and shell input require opt-in, with LAN transport limits.
- Git/PR panels and worker diffs support review while PR AI review is read-only.
- Cross-agent handoff preserves a selected context packet; colleague/runtime takeover is not established.
- Worker artifacts, project panes and local backups retain outputs.
- Playbooks may include checks/conditions, and PR merge checks guard current head.
- Git Control handles status/stage/commit/branches/diffs and worktree PR lookup.
- Squash merge is an explicit human action guarded against stale head/draft/conflicts.
- Clean worker worktree apply and stale-head merge guard are documented.
- Application source is AGPL-3.0-or-later; hosted future services have separate scope.
- Local core requires no Alethe account and has profile backup/settings/cloud export controls.
- Privacy guide distinguishes local diagnostics, default network polls and optional cloud services.
- The desktop is free.
- Early public desktop release has installers and public issue support.
- Free desktop/core supports evaluation with own agents; cloud features are separate.
Sources
Archoninitial researchresearch correction42 findings · 43 sources
Findings today
- Archon has a product CLI and self-hosted browser console.
- Maintained provider plugins run Claude Code, Codex and Pi using their CLIs/SDKs.
- Archon runs existing Claude Code, Codex and Pi accounts.
- Users install a provider, install Archon and run a named workflow in a git repository.
- Archon exposes a REST API under /api with OpenAPI 3 generation.
- Versioned plugins supply workflow packs, agent providers, forges and chat adapters through a published contract and…
- The SDLC pack takes GitHub issues through intake, build and merge queues via the forge plugin.
- YAML workflows declare ordered or dependent agent, script, loop, wait and approval nodes, with output contracts and…
- A workflow node can launch separate child runs with their own identity, artifacts, costs and pauses.
- Archon reuses provider project guidance and stores workflow commands and scripts beside YAML.
- The discoveries workflow reads out-of-task findings and files them as new GitHub issues or attaches them to existing…
- Workflow fan-out creates concurrent governed child runs and max_parallel limits active children.
- The build queue claims ready issues; the merge queue processes ready PRs and returns conflicts for rework.
- The product assistant can invoke workflows and inspect runs; workflow nodes delegate child runs with output returned…
- Native schedules, GitHub events and Standard Webhooks request governed runs with host identity, overlap policy and…
- Console actions include approve, reject, resume, cancel and abandon.
- The console lists runs across projects and exposes per-run event logs, artifacts and graph context; alerts announce…
- Users invoke a ready workflow in natural language or edit the YAML graph to combine agent prompts, scripts, waits,…
- Archon can run locally or be deployed with Docker or on a cloud VPS for always-on operation, using SQLite by default…
- Repository runs receive their own Git worktree by default, permitting separate runs against the same repository.
- Normal container pauses stop compute; resume can restart or recreate the container from its persisted upper volume.
- Container runs require Docker and a current runner image containing tools.
- Optional run budgets enforce work-unit and provider-reported USD spend ceilings before new attempts.
- Run details expose event logs and artifacts; live events show governance state.
- Child run costs roll up into parents without duplicating prior resumed increments; per-run token and provider-cost…
- Trigger bindings declare host, shared resource and overlap policy and share admission records.
- New identities are members.
- Workflow approval gates pause for explicit approve or reject decisions, can rework on rejection and record decisions…
- Each run retains an event log, transcript and cost in the user database.
- Operators open run details with event logs and artifacts and approve, reject, resume, cancel or abandon runs from…
- The console retains workflow events and artifacts, but the scoped product UI has no dedicated proposed-code/diff…
- The SDLC build queue plans, implements, reviews through specialist lenses and fixes findings; its merge gate checks…
- The forge-backed SDLC pack opens PRs, waits for CI and sends conflicting PRs back through build work before merge.
- The merge queue checks a ready review and CI against the head commit and base health before merging.
- SDLC output links the issue and pull request with a ready-review result and exact-head merge checks.
- Build and merge queues coordinate ready issues and PRs, review verdicts, CI and conflict rework.
- Current Archon application source is MIT licensed at the pinned revision.
- Archon is a full self-hosted application on the user machine or server, using SQLite by default or PostgreSQL.
- Archon sends anonymous categorical workflow/platform/provider/outcome and numeric usage events, excluding code and…
- The current core Archon application is free MIT software to install and self-host, with no paid ADE tier in the…
- The repository released v0.14.0 on November 7.
- The core app is available as free MIT software to install and self-host; paid model and own-compute costs remain…
Sources
- Archon: Quickstart; Connect a platform
- Archon: Your agents, your subscriptions; Plugin kinds
- Archon: Your agents, your subscriptions; Quickstart
- Archon: Quickstart
- Archon: API Reference; Authentication
- Archon: A reliable core; Plugin kinds
- Archon: A software factory
- Archon: Workflow language
- Archon: Separate governed child runs; fan_out
- Archon: Your agents; Workflow packs
- Archon: A software factory
- Archon: fan_out; Concurrency
- Archon: A software factory
- Archon: Launching a separate governed run
- Archon: Start workflows from schedules and signed webhooks
- Archon: Workflows and runs; Alerts
- Archon: Console navigation; Workflows and runs
- Archon: Build any factory; Quickstart
- Archon: Deployment overview
- Archon: What it looks like
- Archon: How it works; Provider support; Security posture
- Archon: Pause economics
- Archon: Prerequisites; Running
- Archon: Run budgets; Retry policy
- Archon: Workflows and runs
- Archon: Cost roll-up; Reading tokens and cost
- Archon: Governed admission
- Archon: Users and roles; Who can act on a run
- Archon: Approval nodes
- Archon: Governance built in
- Archon: Workflows and runs
- Archon: Workflows and runs
- Archon: A software factory
- Archon: A software factory
- Archon: The merge queue lands it
- Archon: A software factory
- Archon: A software factory
- Archon: MIT License
- Archon: Self-hosted; Quickstart
- Archon: Telemetry
- Archon: MIT License; Distribution grant
- Archon: v0.14.0 release
- Archon: MIT License
Open-Inspectinitial researchresearch correction37 findings · 20 sources
Findings today
- The primary product UI is a web workspace with sessions, automations, review and settings.
- OpenCode, Claude Agent SDK and Codex are separate selectable harnesses; each session keeps its initial harness for life.
- Claude account works on Claude Agent, ChatGPT account on supported OpenCode/Codex routes, and SuperGrok on supported…
- The setup guide separates running the web app against an existing backend from a full self-hosted deployment, and…
- Control-plane session endpoints include budget updates and operator/service exports, with role/token authorization.
- Managed shared/personal MCP connections are available with harness-specific loading restrictions.
- The client package is Next.js web; no native desktop shell was established.
- Web prompts, Slack, GitHub PRs, Linear issues and webhooks can start coding sessions.
- Sessions retain child/grandchild relationships with a bounded depth and separate sandbox/branch per child.
- Harness-native repository instruction files and pinned managed skills provide session context.
- Parent continues while independent child sandboxes run simultaneously.
- Automation admission rechecks executor/target permission; denied scheduled authorization pauses the automation.
- Child can override model/reasoning but inherits its parents harness and authentication.
- Cron automation starts ordinary sessions without a human prompting each run.
- Authorized collaborators can stop the current session turn.
- Operators self-host the control plane and configure their own remote sandbox-provider accounts for agent execution.
- Each child gets its own branch and sandbox rather than the parents working branch.
- Provider-specific expiry can snapshot/restore or stop/pause/archive and later resume retained state.
- Prebuilt images and saved-state models differ by provider; Modal supports per-turn snapshots while several backends…
- A reported USD session limit stops active work and blocks queued/new prompts until the owner raises/removes it.
- Sessions share token/tool/step/warning events across harnesses; child history can be retrieved programmatically.
- Budget sidebar displays reported session cost and configured limit.
- Child spawning fails when the configured concurrent limit is reached; it is not described as silently queued beyond…
- Participants and connected-presence dots update for viewers in the same session.
- Owner, Administrator, Member and Viewer have documented workspace permissions.
- Credential brokering caches data on disk and snapshots can retain it; GitLab uses a deployment PAT.
- The audit UI records operation/authorization outcomes with actor/resource snapshots; Allowed is not evidence the…
- Changes panel has path filters, unified/split views and Markdown preview with session-start/head context.
- An authorized colleague can open history, queued prompts and current sandbox state and continue the same branch.
- Changes includes committed/uncommitted tracked and nonignored untracked files since session start, with named…
- GitHub PRs and checks are review evidence; a failed-test or requirement-conflict shipping gate was not exercised.
- Each child branches from the parents base branch in its own sandbox.
- Child UI and session review expose linked pull request state and navigation.
- The web/control-plane application source uses MIT.
- Full self-hosted deployment is documented, with the control plane on Cloudflare Workers by default or in a container.
- Snapshots can retain on-disk cached credentials; removing a grant does not instantly revoke issued tokens.
- Public repository, changelog and setup guides provide support context without a response commitment.
Sources
- Open-Inspect README.md
- Open-Inspect docs/SETUP_GUIDE.md
- Open-Inspect packages/docs/content/docs/models/agent-harnesses.mdx
- Open-Inspect packages/docs/content/docs/models/provider-accounts.mdx
- Open-Inspect packages/docs/content/docs/sessions/child-sessions.mdx
- Open-Inspect packages/docs/content/docs/sessions/multiplayer-and-sharing.mdx
- Open-Inspect packages/docs/content/docs/administration/workspace-access.mdx
- Open-Inspect docs/AUTH.md
- Open-Inspect packages/docs/content/docs/configure/sandbox-providers.mdx
- Open-Inspect packages/docs/content/docs/sessions/reviewing-changes.mdx
- Open-Inspect packages/docs/content/docs/sessions/spend-limits.mdx
- Open-Inspect packages/docs/content/docs/integrations/mcp-server.mdx
- Open-Inspect packages/docs/content/docs/configure/mcp-servers.mdx
- Open-Inspect packages/docs/content/docs/configure/memories.mdx
- Open-Inspect packages/docs/content/docs/automations/index.mdx
- Open-Inspect packages/docs/content/docs/administration/data-controls.mdx
- Open-Inspect packages/docs/content/docs/administration/audit-log.mdx
- Open-Inspect packages/docs/content/docs/reference/limits.mdx
- Open-Inspect LICENSE
- Open-Inspect runtime architecture
bbinitial researchresearch correction35 findings · 17 sources
Findings today
- Desktop supports macOS arm64; Linux x64 and Windows x64 installers are alpha.
- Installed Claude Code, Codex, Pi, Cursor via ACP, OpenCode, Grok Build and other compatible ACP agents are documented.
- Codex accepts ChatGPT login, while Claude uses the signed-in CLI and reports its plan.
- Install/download bb, authenticate at least one agent provider and open the served app.
- bb CLI exposes thread and project operations; plugin commands extend it.
- Built-in provider, task, automation, memory, capacity and MCP plugins extend bb.
- The repository documents an Electron desktop shell around the web app.
- Tasks plugin creates product-owned task records independent of running threads.
- Task descriptions, subtasks, comments and attachments supply durable worker briefs.
- The Tasks plugin supports subtasks and nested folders; arbitrary blocker enforcement remains unverified.
- Global/project memory has provenance, kinds, tags, versions and a summary catalog injected each turn.
- Product-owned tasks support create/edit, board/list views, filters and task detail.
- bb tasks CLI reads and writes the same stable task keys shown to humans.
- Managed worktrees allow concurrent threads without changing the main checkout.
- Task delegation selects provider/model/effort/permissions and attaches a worker thread.
- Manager threads coordinate child threads; provider-native subagents can be disabled to route delegation through bb.
- Automations support repeating cron/timezone and one-shot schedules.
- Live threads can receive steering through browser, CLI and HTTP/SDK surfaces.
- Default launcher starts server and local host daemon on the current machine.
- Managed worktrees separate checkout directories/branches.
- The launcher restarts crashed server/daemon child processes separately.
- Per-worktree setup scripts install dependencies before dispatch; failure or 15-minute timeout blocks thread start.
- The changelog describes failing provider turns accepted but not started after five minutes.
- Service logs rotate and SQLite thread events retain timeline progress.
- New idle-thread turns queue at host/global capacity and recheck after another thread settles.
- Approved remote MCP app grants can be revoked.
- Threads retain an append-only stream of messages, tool calls and file changes in server state.
- The Git Diff tab displays uncommitted, branch and per-commit diffs with Viewed marks and filters.
- Managed worktree environments create separate branches for parallel work.
- PR merge actions are provided by the GitHub PR status plugin; exact-revision authority enforcement is unverified.
- Thread Info retains the linked pull request even when the above-composer PR panel is hidden.
- bb application source is MIT licensed.
- The central server, SQLite state and execution daemons can run on user-controlled machines.
- Optional storage-retention plugin sets archive/delete policies and defaults to disabled.
- README calls the project active development and Linux/Windows desktop alpha.
Sources
- bb README.md
- bb packages/bb-app/README.md
- bb docs/system-overview.md
- bb docs/multiple-devices.md
- bb docs/platform-support.md
- bb docs/worktrees.md
- bb plugins/tasks/PLUGIN_OVERVIEW.md
- bb plugins/automations/PLUGIN_OVERVIEW.md
- bb plugins/concurrency-limit/PLUGIN_OVERVIEW.md
- bb plugins/memory/PLUGIN_OVERVIEW.md
- bb plugins/mcp/PLUGIN_OVERVIEW.md
- bb plugins/provider-claude-code/PLUGIN_OVERVIEW.md
- bb plugins/provider-codex/PLUGIN_OVERVIEW.md
- bb docs/configuration.md
- bb plugins/git/PLUGIN_OVERVIEW.md
- bb CHANGELOG.md
- bb LICENSE
Boltinitial researchresearch correction37 findings · 39 sources
Findings today
- Bolt works in Chromium desktop browsers and Safari.
- Bolt supplies Standard and Max agents, plus Forge for eligible accounts.
- Bolt subscriptions provide token allocations for builder usage; Free use refreshes daily and monthly and paid…
- Users start with prompts, Figma, Stitch, GitHub or a Lovable import.
- A project owner can connect built-in or custom MCP servers to Bolt and authorize external tools and data.
- A natural-language prompt creates a project; GitHub and design imports bring an existing starting point.
- Plan Mode lets users discuss an approach before code changes, then switch to building after refining scope.
- Project settings store project knowledge for the agent, separate from a single chat prompt.
- Code view permits browsing files and paths for references in prompts.
- A Bolt project has one shared chat and processes one prompt at a time; prompts sent during generation are queued.
- Bolt queues later prompts from collaborators until the current project prompt completes.
- Chat tools let users queue follow-up prompts while Bolt is busy and manage the queued requests.
- Users can build and refine with natural-language chat, change agent choice and optionally edit source directly.
- The development environment runs on WebContainers, which run Node.js in the browser.
- Bolt can create and switch GitHub branches and keeps branch-specific memory.
- Built-in version history lets users restore earlier project versions.
- Chromium Code view exposes a terminal to run project commands; Safari does not expose the terminal.
- Free use has daily and monthly limits.
- The GitHub panel shows Syncing or Synced status, linked repository and active branch.
- Builder usage is measured in tokens and visible in account usage.
- Shared-project prompts run serially and excess prompts queue.
- Collaborators prompt and edit the same project in one shared chat.
- Team Admins manage membership, roles, billing, project controls and integrations; Members build within those controls.
- Only the project owner manages the GitHub connection.
- Project version history and restore preserve prior code states.
- Users can leave threaded feedback on preview elements.
- Collaborators share a single chat and project, while GitHub sync waits for the original project owner to return.
- Bolt exposes a live app preview and source through Code view.
- Users can run a project security audit on paid plans or a database security check on all plans, review the findings…
- Bolt creates and switches branches and automatically syncs project changes to GitHub.
- Bolt does not currently merge branches in-app; users merge through GitHub.
- Users can restore prior project versions and export project backups.
- Project code can be downloaded as a backup and synced to GitHub.
- Bolt offers Free, USD 25/month Pro, USD 30/member/month Teams and custom Enterprise at monthly billing.
- Release notes document database tables becoming private by default starting October 30 and expanded phone browser…
- The ongoing Free plan supports build, publish, viewer/editor/co-owner project sharing and recurring daily/monthly use.
- Plans disclose token allocation, monthly or annual billing and a downgrade to Free when paid cancellation takes effect.
Sources
- Bolt: Supported browsers; Mobile
- Bolt: Agents
- Bolt: Billing
- Bolt: Ways to start a project; Bolt and StackBlitz
- Bolt: Connect to an MCP server
- Bolt: Ways to start a project
- Bolt: Plan before building
- Bolt: Project knowledge
- Bolt: Search; Reference files
- Bolt: Work together in real time
- Bolt: Work together in real time
- Bolt: Queue prompts
- Bolt: Intro to Bolt
- Bolt: Bolt and StackBlitz
- Bolt: Safari
- Bolt: Branching and merging
- Bolt: Version history
- Bolt: Use the terminal
- Bolt: What is included in the Free plan
- Bolt: Check your sync status
- Bolt: Tokens
- Bolt: Work together in real time
- Bolt: Work together in real time
- Bolt: Team member roles
- Bolt: Share integrations
- Bolt: Version history
- Bolt: Add comments to your project and send them to Bolt
- Bolt: Work together in real time; Share integrations
- Bolt: Intro to Bolt
- Bolt: Security audit
- Bolt: Branching and merging; GitHub sync
- Bolt: Change branches in Bolt
- Bolt: Project and Site
- Bolt: Backups, restore and version history
- Bolt: Export your project
- Bolt: Billing; Monthly plan example
- Bolt: October 26 to November 1
- Bolt: What is included in the Free plan
- Bolt: Billing; Cancel your subscription
claude-code-actioninitial research25 findings · 10 sources
Findings today
- The product is a GitHub composite Action.
- The action runs the Claude Code harness through its SDK run path.
- Claude Pro/Max users can generate a Claude Code OAuth token using claude setup-token for the action.
- A repository admin installs the GitHub App, stores credentials and configures the workflow.
- The workflow accepts prompt/claude_args and emits structured JSON action outputs.
- Custom --mcp-config servers merge with built-in GitHub MCP servers; allowedTools gates enabled tools.
- Issue and PR comments or an explicit workflow prompt supply task input.
- README links scheduled maintenance examples using GitHub Actions.
- The FAQ describes using GitHub Actions logs and optional debug output for diagnostics.
- The action runs on the selected user-controlled GitHub runner.
- The action expects a Git repository at the runner workspace root.
- claude_args supports --max-turns to bound a Claude invocation.
- GitHub Actions logs expose execution diagnostics; full output may contain sensitive content.
- Repository write access normally gates supported actor-triggered events; documented allowed_non_write_users and bot…
- allowedTools/disallowedTools configure Claude tool access, and GitHub token permissions bound API operations.
- Interactive mode updates a single GitHub tracking comment with progress and final output.
- Validated JSON results become workflow action outputs.
- The action can retrieve configured CI evidence; enforcing branch protection remains a separate GitHub policy.
- Default issue mentions create a work branch; existing open PR work pushes commits to that PR branch.
- The default action documentation says it does not merge branches; this is described as a prompt restriction.
- The default result includes a human PR-creation link, distinct from evidence that a PR was created.
- The action repository code is MIT licensed.
- The core action executes on the users own GitHub runner, still using GitHub and the chosen model provider.
- Full debug output may include sensitive context, tokens and tool content and requires access restrictions.
- The latest retrieved version is v1.0.269 dated 2026-11-06.
Sources
- Claude Code Action README.md
- Claude Code Action docs/setup.md
- Claude Code Action docs/usage.md
- Claude Code Action docs/configuration.md
- Claude Code Action docs/security.md
- Claude Code Action docs/faq.md
- Claude Code Action docs/cloud-providers.md
- Claude Code Action docs/custom-automations.md
- Claude Code Action LICENSE
- Claude Code Action v1.0.269
claude_codex_bridgeinitial researchresearch correction28 findings · 5 sources
Findings today
- CCB provides a native terminal application for Linux/macOS and a publicly downloadable native Windows x64 beta.
- Named CLI families include Claude Code, Codex, Gemini, OpenCode, Pi, Grok, Cursor and additional providers.
- Release history describes preserving local conversation history across credential/key/gateway changes for Claude/Codex.
- Install @seemseam/ccb, run ccb in a project, then configure installed providers using local Settings or ccb config ui.
- ccb config validate/update and ask manage topology and inter-agent collaboration.
- Optional Agent Roles packs bundle skills, memory and tools and are installed separately.
- CCB Mobile is a Flutter Android app distributed as an APK.
- A new project opens one demo agent using the first available CLI; a full team is not mounted by default.
- Project-wide .ccb/ccb_memory.md stores shared collaboration rules and durable context.
- Configured windows can run multiple CLI providers concurrently.
- ask requests and returned results share a per-target chronological FIFO.
- The Config UI edits per-agent provider, model, thinking level, pane topology and worktree scope.
- Humans can type in native panes or send phone input; queued messages wait for an eligible idle input surface.
- A project starts with one agent and can grow through visual pane/topology configuration and guarded reload.
- CCB runs installed provider CLIs and a project daemon on the current host.
- CCB separates configured workers with Git worktrees and terminal processes.
- v8.7.21 uses private legacy OpenCode database snapshots and opt-in v2 native ingress; uncertain historical jobs are…
- Managed draft protection waits up to 180 seconds before clearing a nonempty draft once; busy/modal states keep waiting.
- ccb-diagnose combines authoritative job state, queue/inbox/trace/provider logs and live-pane evidence for a named agent.
- One target agent processes ordered requests/results serially while different agents can work in parallel.
- Pairing profiles authorize view/content/terminal/file-transfer scopes; exact runtime authorization limits require…
- Project daemon state and shared memory survive closing the foreground UI; no immutable audit-retention claim is made.
- Optional Rich mode browses files, edits documents and previews media inside WezTerm/Yazi.
- Explicit worker(worktree) topology separates worker checkouts.
- CCB declares AGPL-3.0; its own license summary also discusses separately licensed closed-source commercial use.
- Core TUI/daemon and mobile gateway run on user-controlled machines from the distributed source/package.
- Version 8.7.21 is current; native Windows x64 remains beta and some provider ingress/queue behavior is opt-in.
- The license summary explicitly states CCB can be used for free, subject to its licensing terms and separate model costs.
Sources
CodeNomadinitial researchfreshness recheck34 findings · 8 sources
Findings today
- CodeNomad supplies native desktop packages for macOS, Windows and Linux and a standalone browser/PWA workspace.
- CodeNomad's built-in coding runtime is OpenCode V2.
- OpenCode-native OAuth subscriptions and API credentials are supported, with account selection scoped service-wide.
- Desktop includes Node/npm; select or install OpenCode, open a folder and connect a provider.
- Product-owned CLI launches and configures the standalone server.
- Workspace manages MCP connections, Global/Project plugins and SideCar web tools.
- Desktop variants use Electron or Tauri, with a SolidJS frontend and Node server.
- Native missions save an objective and task tree separately from conversations, and create remains paused until…
- Missions retain durable plans, task briefs and reports with explicit amendment controls.
- Mission tasks record dependencies, execution identities and nested conversations.
- Project files, saved briefs, playbooks and followed conversation context feed missions.
- File references, tool images and agent-driven web previews are documented.
- Missions have durable tasks with states, briefs and reports, separate from session lists.
- Mission coordinator plans and delegates into native conversations and records task results.
- Mission objective and notes are durable, editable and linked to tasks; formal requirements approval remains…
- Multiple projects, sessions and specialist conversations can run concurrently.
- Mission coordinator owns task assignment and preserves original control identities.
- Product mission coordinator plans, delegates and exposes specialists, reports and progress in an inspectable tree.
- Native schedules and event launch criteria can arm future mission passages.
- Pause, Interrupt now, Stop and uncertainty-aware refresh are distinct controls; native permission questions appear…
- Mission task tree exposes dependency links, progress and retained reports.
- Five playbooks provide flexible mission, debugging, exploration, product delivery and two-model comparison.
- Standalone server executes on user-controlled host and can continue while the phone PWA is closed; host must remain…
- CodeNomad supplies worktree file separation.
- Shared OpenCode service survives closing CodeNomad clients; reconnect refetches state.
- Mission status distinguishes admitted input from observed work.
- Per-message tokens and cost estimates are separate from provider quota and wallet readings.
- Native OpenCode permission controls remain in force; the browser UI does not authorize all later tool effects.
- Conversation diffs, project files, tool images, previews and mission reports stay linked to work.
- CodeNomad manages Git/worktrees and can list forge issues and pull requests read-only.
- CodeNomad application source is MIT licensed.
- Core server and clients run on user-controlled hardware without CodeNomad accounts.
- CodeNomad says it has no telemetry or analytics.
- Desktop/browser source pass covers current 1.0 and OpenCode V2, including dev releases; install and upgrade…
Sources
codex-actioninitial researchfreshness recheck26 findings · 4 sources
Findings today
- The product is a composite GitHub Action with no standalone desktop, browser or phone client.
- The action installs and invokes only Codex CLI; endpoint customization changes the model provider, not the coding…
- Provider API key is required; a ChatGPT subscription-only route is not documented.
- Workflow must provide checkout, dependencies and secrets before invoking the action.
- The bundled codex-action helper CLI provides setup, permission checks, proxy configuration and run-codex-exec…
- Trusted Codex config can attach MCP tools; the action does not own an MCP service.
- GitHub event context and prompt text are workflow-authored inputs.
- Prompt files can hold task instructions; no durable native planning entity was established.
- Underlying Codex can load repository instruction files and trusted configuration.
- GitHub Actions owns event/schedule dispatch around this action.
- Final-message output and GitHub job logs expose result; interactive steering is not established.
- Codex runs on the caller-selected GitHub Actions runner.
- The action configures Codex permission profiles and runner privileges; it does not provide worktree, container or VM…
- Startup validates conflicting permission inputs and protected flags; task budget/retry behavior was not tested.
- GitHub workflow logs and final-message expose the action result.
- The action checks GitHub repository write permission and caller allowlists; it supplies neither native workspace…
- Privilege reduction and command permission profiles are explicit configuration controls.
- The action returns final-message output or an output file.
- Caller controls checks and follow-on jobs; no action-native release gate was established.
- The documented review bot returns Codex output to a caller-authored GitHub comment job.
- GitHub workflow author grants delivery authority through later jobs and tokens.
- Action source is Apache-2.0 licensed.
- The product code and execution worker can run on user-controlled runners, but GitHub Actions supplies workflow…
- The action manages provider credentials and streams Codex output.
- Public action source and issue tracker exist, without a separate release or support commitment established.
- Apache-licensed action source is available; execution incurs separate API/runner costs.
Sources
collaboratorinitial researchfreshness recheck13 findings · 3 sources
Findings today
- Collaborator ships a native Electron desktop app for macOS, Windows and Linux.
- Open a local folder, create a terminal tile and start an installed agent.
- Desktop app uses Electron 40, React 19, xterm.js, Monaco and a persistent node-pty sidecar.
- Local context files can sit beside terminals as editable Markdown or code tiles; managed shared-memory provenance is…
- Independent terminal sessions can run agents side by side on the canvas.
- The scoped Electron app runs local terminal processes on the user computer.
- Terminal sessions persist independently of the canvas tile lifecycle, and canvas/config state is written to disk.
- Terminal tiles open in the workspace directory.
- The canvas displays editable code and file tiles; its source does not implement a proposed-change diff review interface.
- Collaborator application source uses FSL-1.1-ALv2, with a future Apache-2.0 conversion.
- Core application can be built and run locally, with no product account required.
- The application collects anonymous usage events with PostHog.
- Early-stage desktop app; latest listed release 0.8.4 was published June 16, 2026.
Sources
Zeroninitial researchresearch correction28 findings · 18 sources
Findings today
- The pinned release supplies native desktop and phone clients plus a headless engine; it does not supply a released…
- Zeron runs the documented built-in harnesses plus registry-provided ACP agents.
- Cloud credentials support native Claude OAuth/setup-token and ChatGPT/Codex OAuth, with scope limited to the…
- Download a desktop build or install the Linux headless service; authenticate the chosen agent separately.
- zeron status/update/daemon and mcp are product-owned commands.
- The engine injects its control MCP into supported harnesses, separate from provider-native subagents.
- Desktop is Rust/GPUI around a Rust engine, replacing the older Electron viewport.
- Users create chats in local/synced spaces; a durable chat is not evidence of separate product issue tracking.
- Product-owned side chats retain a parent link, limited to one level; children cannot create more chats.
- Pi discovers commands/skills in the selected workspace and controls project-extension trust.
- MCP create_chats supports multiple children and Cloud can provision per-session machines.
- Messages carry origin chat identity and self-messaging is refused.
- Shared queue distinguishes step-boundary steering, turn-end waiting and Send now interruption according to harness…
- Local execution remains the default owned-device mode.
- Engine manages Git worktrees as separate checkouts.
- The engine retries unexpected daemon loss with bounded backoff and stops after repeated loss; recovery is not…
- Explicitly imported project Actions may run setup on new worktrees; reusing an existing checkout does not repeat setup.
- Background tray reads bounded live command output and waits for runtime confirmation before removing stopped entries.
- Only the host drains durable queues; edit/review gates block delivery instead of skipping the head.
- Project action imports require explicit approval; opening a repository does not authorize its suggested shell commands.
- Durable command ledger and queue rows retain stable IDs, authors/timestamps and host outcomes.
- Native Changes viewer renders file/hunk/line patches with comparison and error states.
- Worktree RPCs and project actions manage separate checkouts.
- Zeron source is MIT licensed.
- Core engine and desktop run locally or on an owned Linux server; a supported self-hosted sync backend contract is…
- Native apps have no analytics/crash SDKs, and the pinned app settings/engine exposes no app telemetry opt-out control.
- Latest retrieved release 0.2.139-beta.2 is a next-app beta with a separate beta update channel.
- The Zeron desktop app is a free MIT download from GitHub Releases with no account needed, and the iPhone/iPad…
Sources
- Zeron README.md
- Zeron ARCHITECTURE.md
- Zeron docs/PARITY.md
- Zeron docs/mcp.md
- Zeron docs/design/cloud-device.md
- Zeron docs/pi.md
- Zeron docs/reference/message-queue.md
- Zeron docs/reference/project-actions.md
- Zeron docs/program-status.md
- Zeron docs/session-publication.md
- Zeron docs/background-processes.md
- Zeron apps/ios/README.md
- Zeron apps/android/README.md
- Zeron LICENSE
- Zeron official website
- Zeron developer App Store listing
- Zeron privacy policy
- Zeron beta-v0.2.139-beta.2
dmuxinitial researchfreshness recheck19 findings · 3 sources
Findings today
- dmux supplies a terminal CLI/TUI inside tmux on macOS and Linux, not a separate desktop GUI.
- The fixed launch registry supports Claude Code, Codex, OpenCode, Grok Build, Cline, Gemini CLI, Qwen, Amp, Pi,…
- Optional internal inference supports API keys, OpenAI-compatible endpoints, Codex ChatGPT subscription login or Grok…
- Install npm package, enter repository and run dmux; dependencies include tmux, Node, Git and chosen coding CLI.
- dmux owns the CLI and terminal commands for pane/workspace management.
- Lifecycle hooks can run scripts around worktree creation and merge.
- Several coding agents run in parallel with individual worktrees/panes.
- Users jump between panes to interact, receive macOS attention alerts, and can hide panes without stopping work.
- Prompt intake can select multiple agents, optional goal mode and custom branch/base names.
- Agents run in tmux on the machine where dmux is invoked.
- dmux creates a branch and Git worktree per agent pane.
- Regular terminals restore last observed directory; tracked coding agents resume conversations when panes are recreated.
- Lifecycle hooks can prepare worktrees and handle merge steps; automatic dependency provisioning unestablished.
- Built-in file browser searches worktree files and previews code/diffs.
- Pane menu can merge completed work or push branch and create GitHub PR; custom base/branch names are supported.
- Merge menu brings task branch back to main and cleans it up.
- dmux application source uses MIT license.
- Core dmux CLI runs on user-controlled hardware with local tmux/Git/Node and installed agents.
- Latest listed release v5.11.1 was published August 16, with terminal naming/activity improvements.
Sources
Fusioninitial researchfreshness recheck40 findings · 8 sources
Findings today
- Web dashboard (installable PWA), CLI, a released Windows x64 Electron desktop app and an Android APK.
- Fusion has explicit installable Cursor Agent and Grok CLI runtime plugins, plus an experimental generic ACP runtime…
- Subscription OAuth, API-key and local-model paths are separately documented.
- One-command daemon and dismissible onboarding support local setup; remote use requires a running host.
- Product CLI and HTTP APIs create/manage tasks, agents and routines.
- Fusion owns MCP configuration and plugin setup; actual tool calls depend on runtime support.
- Desktop is Electron and mobile is Capacitor, with a separate browser dashboard.
- Native tasks are created in dashboard/CLI or imported from GitHub; external sync scope is not assumed.
- Planner writes PROMPT.md with steps, file scope and acceptance criteria.
- Mission hierarchy and dependency-aware dispatch are documented.
- Agents retain project and identity instructions/memory; no product-spec drift fixture was run.
- Research and knowledge tools retrieve web/GitHub/local documents; architectural enforcement remains untested.
- Product-owned tasks have durable state, metadata and board/list/graph views.
- Agent tools create, assign and delegate native task records.
- Independent tasks execute concurrently with configurable project/system limits.
- Durable agent roles, assignment tools and opt-in backlog claims route work.
- Fusion owns orchestration, task delegation and agent mail.
- Daemon schedules and webhook routines start work unattended.
- Active task details expose nudge/pause/re-prompt and question response cards.
- Boards, graphs, mission planner and Command Center expose multiple work items.
- Quick fix, coding, review-heavy and custom graph workflows are selectable per task.
- Fusion runs its engine/dashboard on user-controlled local or remote hosts.
- Tasks use Git worktrees with optional native, bubblewrap, macOS Seatbelt, Docker/Podman or custom execution backends.
- Chat streams are resumable and remote clients share daemon state; crash recovery was not tested.
- Step timeouts, task token budgets and retry/escalation settings are documented.
- Command Center, task logs and task chat expose usage and progress.
- Costs are estimates based on provider-qualified model prices; unknown pairs remain unavailable.
- Engine caps project/system agents, task claims and process-wide verification.
- Bearer access, trusted MCP and planner confirmation policy provide distinct documented boundaries.
- Task detail combines plan, reviews, diffs and live file changes.
- Tasks retain plans, reviews, diffs, files and attachments.
- Configurable workflows distinguish pre-merge blocking gates and informational post-merge steps.
- Integrated Git worktrees, PR creation and GitHub issue import are documented.
- Merge behavior is configurable; planner confirmation claims coexist with global/per-task auto-merge.
- Task workflow links completion to direct squash merge or merged PR.
- Review columns and PR workflow fragments organize several changes, without a tested six-task review fixture.
- Fusion repository is MIT licensed; external provider terms remain separate.
- Core daemon, dashboard and clients are self-hostable.
- Fusion is early preview with beta releases and public support channels.
- Open-source local install offers evaluation with own providers and compute.
Sources
Gas Towninitial researchresearch correction41 findings · 42 sources
Findings today
- Gas Town supplies a gt CLI and browser dashboard.
- Gas Town supplies built-in runtime presets and generic user-defined commands.
- Users install the underlying agent CLI and authenticate it themselves.
- A native host requires Git, Beads, Dolt and tmux for the full agent roles; Docker packages prerequisites.
- The product-owned gt CLI manages workspaces, agents, dispatch, convoys, scheduler and escalation; the browser…
- TOML formulas define reusable work steps, and custom runtime aliases configure external coding agents.
- Gas Town tracks work as Beads issues with stable prefix IDs, groups them in convoys and dispatches by issue ID.
- TOML formulas are frozen into templates and instantiated as tracked work.
- Dispatch can target an epic and its children or a convoy and tracked issues.
- gt prime reconstructs role/work context and gt seance finds earlier sessions or queries a predecessor.
- The bundled Beads ledger gives persistent issue IDs and work state; the dashboard displays issues, hooks, convoys…
- Agents use gt/bd commands to inspect hooks, instantiate work, assign issue IDs, report done and escalate blockers in…
- Gas Town manages multiple coding agents working on different tasks.
- gt sling either dispatches directly or creates a separate ephemeral sling-context bead.
- The Mayor is a product-configured AI coordinator with project/agent context.
- The daemon heartbeat auto-dispatches bundled or user plugins with an elapsed cooldown gate to idle dog agents, and…
- The activity feed exposes stuck, stalled, zombie, working and idle agents.
- The terminal feed groups agents by rig/role, lists in-progress/recent convoys and streams events.
- Users can tell the Mayor what to build, manually sling issue IDs or run minimal tracking mode with their own agent…
- The full workspace manager runs on the user-selected host, including a server reached by terminal or the detached…
- Git worktrees and optional Docker provide the scoped execution separation.
- Poured workflows retain completed step rows and resume from the last checkpoint after a session dies.
- The runtime image bundles prerequisites.
- Enabled deferred dispatch controls concurrency to avoid rate-limit exhaustion.
- OpenTelemetry emits structured session, dispatch, agent-state, mail, formula and daemon events.
- The configured scheduler caps polecats, queues work and respects readiness before dispatch.
- The Copilot autonomous preset uses --yolo and Kiro uses --trust-all-tools.
- Agent identities, work provenance and actions are tracked in structured ledger records.
- The dashboard exposes completed work, merge queues and verification status, but does not display proposed-code diffs…
- The federation workflow accepts completion evidence URLs tied to wanted-work IDs.
- The Refinery batches merge requests, runs verification on the stack and bisects a failing batch to merge only…
- gt done pushes a worker branch and creates a merge-request bead for the internal Refinery.
- Workers do not push directly to the integration branch; the Refinery merges passing batches and isolates failed ones.
- Completed work is tracked through merge-request beads and hooks can be archived.
- The Refinery implements a bisecting queue for batches of worker changes, with verification and failed-work redispatch.
- Gas Town source is MIT licensed at the pinned default-branch revision.
- The CLI, ledger, daemon and dashboard run on the user host or Docker runtime.
- Telemetry is off unless an operator sets a GT_OTEL endpoint; events are then exported over OTLP and correlated by…
- The core software license permits free-of-charge use.
- The latest returned release is v1.2.1 dated June 6, while the pinned default branch changed October 28.
- The core application is available as free MIT software for self-hosting.
Sources
- Gas Town: Installation; Dashboard; Activity feed
- Gas Town: Runtime Configuration; Agent Operations
- Gas Town: Prerequisites; Runtime Configuration
- Gas Town: Installation; Create your workspace
- Gas Town: Key commands; Dashboard
- Gas Town: Cooking Formulas; Runtime Configuration
- Gas Town: Beads Integration; Manual Convoy Workflow
- Gas Town: Molecule lifecycle; How agents see steps
- Gas Town: Quick start; Dispatch cycle
- Gas Town: Seance; Context recovery
- Gas Town: Beads Integration; Dashboard
- Gas Town: Agent operations; Beads integration
- Gas Town: Overview; Scheduler
- Gas Town: Dispatch modes; Sling context beads
- Gas Town: The Mayor; Monitoring; Basic workflow
- Gas Town: Overview
- Gas Town: Problems View; Escalation
- Gas Town: Activity Feed; Dashboard
- Gas Town: Mayor Workflow; Minimal Mode
- Gas Town: Installation; Docker Compose setup
- Gas Town: Hooks; Architecture
- Gas Town: Runtime image; Security posture
- Gas Town: Poured wisps
- Gas Town: Image contents; Storage
- Gas Town: Overview; Dispatch modes
- Gas Town: Telemetry; Problems View
- Gas Town: Overview; Quick Start
- Gas Town: Runtime Configuration
- Gas Town: Why Gas Town exists
- Gas Town: Merge Queue; Activity Feed
- Gas Town: Wasteland federation
- Gas Town: Merge Queue
- Gas Town: Merge Queue
- Gas Town: Merge Queue
- Gas Town: Merge Queue; Hook lifecycle
- Gas Town: Merge Queue
- Gas Town: MIT License
- Gas Town: Installation; Docker setup
- Gas Town: OTel events; Agent content
- Gas Town: MIT License
- Gas Town: Release and maintenance
- Gas Town: MIT License
Archived event links
- 2026-11-13 · Agent Deck
- 2026-11-13 · Agent of Empires
- 2026-11-13 · Orchestrator.inc
- 2026-11-13 · Agent Teams
- 2026-11-13 · ai-devkit
- 2026-11-13 · Alethe
- 2026-11-13 · Amp
- 2026-11-13 · Google Antigravity
- 2026-11-13 · Archon
- 2026-11-13 · Atlas
- 2026-11-13 · Augment Code
- 2026-11-13 · Open-Inspect
- 2026-11-13 · bb
- 2026-11-13 · Bolt
- 2026-11-13 · Buzz by Block
- 2026-11-13 · Capy
- 2026-11-13 · Cate
- 2026-11-13 · Claude Code
- 2026-11-13 · claude-code-action
- 2026-11-13 · claude_codex_bridge
- 2026-11-13 · Claude Squad
- 2026-11-13 · Cline
- 2026-11-13 · cmux
- 2026-11-13 · CodeLayer / HumanLayer
- 2026-11-13 · CodeNomad
- 2026-11-13 · OpenAI Codex
- 2026-11-13 · codex-action
- 2026-11-13 · collaborator
- 2026-11-13 · Zeron
- 2026-11-13 · Conductor
- 2026-11-13 · Cursor
- 2026-11-13 · Devin
- 2026-11-13 · dmux
- 2026-11-13 · Emdash
- 2026-11-13 · Factory
- 2026-11-13 · Fusion
- 2026-11-13 · Gas Town
- 2026-11-13 · gh-aw
- 2026-11-13 · Ghostex
- 2026-11-13 · GitHub Copilot
- 2026-11-13 · Happy
- 2026-11-13 · Herdr
- 2026-11-13 · Intent by Augment
- 2026-11-13 · Jcode
- 2026-11-13 · Jean
- 2026-11-13 · JetBrains Air
- 2026-11-13 · Jules
- 2026-11-13 · Kiro
- 2026-11-13 · Lanes Desktop
- 2026-11-13 · Linzumi
- 2026-11-13 · lody
- 2026-11-13 · loki-mode
- 2026-11-13 · Lovable
- 2026-11-13 · Maestro
- 2026-11-13 · MonoCode
- 2026-11-13 · multi-agent-shogun
- 2026-11-13 · Multica
- 2026-11-13 · Nimbalyst
- 2026-11-13 · Omnara
- 2026-11-13 · Omnigent
- 2026-11-13 · Ona
- 2026-11-13 · open-swe
- 2026-11-13 · OpenChamber
- 2026-11-13 · OpenCode
- 2026-11-13 · OpenHands
- 2026-11-13 · OpenSpec
- 2026-11-13 · Orca
- 2026-11-13 · Paseo
- 2026-11-13 · Podium
- 2026-11-13 · Proliferate
- 2026-11-13 · QM from YC
- 2026-11-13 · ralph-claude-code
- 2026-11-13 · ralph-orchestrator
- 2026-11-13 · Ralph TUI
- 2026-11-13 · ralphex
- 2026-11-13 · Replicas
- 2026-11-13 · Replit
- 2026-11-13 · Rookery
- 2026-11-13 · scion
- 2026-11-13 · Solenta
- 2026-11-13 · Supacode
- 2026-11-13 · super.engineering
- 2026-11-13 · Superconductor
- 2026-11-13 · SuperPlane
- 2026-11-13 · Paradev
- 2026-11-13 · Synara
- 2026-11-13 · T3 Code
- 2026-11-13 · Tembo
- 2026-11-13 · ThinkRail
- 2026-11-13 · Traycer
- 2026-11-13 · Unpeel
- 2026-11-13 · v0
- 2026-11-13 · Verdent
- 2026-11-13 · Vibe Kanban
- 2026-11-13 · Waku
- 2026-11-13 · Warp
- 2026-11-13 · Warp Factories / Oz
- 2026-11-13 · workmux
- 2026-11-13 · Xum
- 2026-11-13 · Zed
- 2026-11-13 · Delta
- 2026-11-13 · Zencoder
- 2026-11-13 · Zenflow Code by Zencoder
- 2026-11-12 · Atlas
- 2026-11-12 · Cate
- 2026-11-12 · SuperPlane
- 2026-11-12 · ThinkRail
- 2026-11-12 · workmux
- 2026-11-11 · Orchestrator.inc
- 2026-11-11 · Herdr
- 2026-11-11 · Lanes Desktop
- 2026-11-11 · Orca
- 2026-11-11 · Paseo
- 2026-11-11 · Podium
- 2026-11-10 · Agent Deck
- 2026-11-10 · Agent of Empires
- 2026-11-10 · Orchestrator.inc
- 2026-11-10 · Agent Teams
- 2026-11-10 · ai-devkit
- 2026-11-10 · Alethe
- 2026-11-10 · Archon
- 2026-11-10 · Open-Inspect
- 2026-11-10 · bb
- 2026-11-10 · Buzz by Block
- 2026-11-10 · claude-code-action
- 2026-11-10 · claude_codex_bridge
- 2026-11-10 · Claude Squad
- 2026-11-10 · Cline
- 2026-11-10 · cmux
- 2026-11-10 · CodeLayer / HumanLayer
- 2026-11-10 · CodeNomad
- 2026-11-10 · OpenAI Codex
- 2026-11-10 · codex-action
- 2026-11-10 · collaborator
- 2026-11-10 · Zeron
- 2026-11-10 · dmux
- 2026-11-10 · Emdash
- 2026-11-10 · Fusion
- 2026-11-10 · Gas Town
- 2026-11-10 · gh-aw
- 2026-11-10 · Ghostex
- 2026-11-10 · Happy
- 2026-11-10 · Herdr
- 2026-11-10 · Intent by Augment
- 2026-11-10 · Jcode
- 2026-11-10 · Jean
- 2026-11-10 · JetBrains Air
- 2026-11-10 · loki-mode
- 2026-11-10 · Maestro
- 2026-11-10 · MonoCode
- 2026-11-10 · multi-agent-shogun
- 2026-11-10 · Multica
- 2026-11-10 · Nimbalyst
- 2026-11-10 · Omnara
- 2026-11-10 · Omnigent
- 2026-11-10 · open-swe
- 2026-11-10 · OpenChamber
- 2026-11-10 · OpenHands
- 2026-11-10 · OpenSpec
- 2026-11-10 · Orca
- 2026-11-10 · Paseo
- 2026-11-10 · Proliferate
- 2026-11-10 · QM from YC
- 2026-11-10 · ralph-claude-code
- 2026-11-10 · ralph-orchestrator
- 2026-11-10 · Ralph TUI
- 2026-11-10 · ralphex
- 2026-11-10 · Rookery
- 2026-11-10 · scion
- 2026-11-10 · Solenta
- 2026-11-10 · Supacode
- 2026-11-10 · Synara
- 2026-11-10 · T3 Code
- 2026-11-10 · Tembo
- 2026-11-10 · Traycer
- 2026-11-10 · Unpeel
- 2026-11-10 · Vibe Kanban
- 2026-11-10 · Waku
- 2026-11-10 · Warp
- 2026-11-10 · Xum
- 2026-11-08 · gh-aw
- 2026-11-08 · Ghostex
- 2026-11-08 · Happy
- 2026-11-08 · Herdr
- 2026-11-08 · Jcode
- 2026-11-08 · loki-mode
- 2026-11-08 · Lovable
- 2026-11-08 · MonoCode
- 2026-11-08 · multi-agent-shogun
- 2026-11-08 · Nimbalyst
- 2026-11-08 · open-swe
- 2026-11-08 · OpenChamber
- 2026-11-08 · ralph-claude-code
- 2026-11-08 · ralph-orchestrator
- 2026-11-08 · Ralph TUI
- 2026-11-08 · ralphex
- 2026-11-08 · Replit
- 2026-11-08 · scion
- 2026-11-08 · Supacode
- 2026-11-08 · Traycer
- 2026-11-08 · v0
- 2026-11-08 · Waku
- 2026-11-07 · Orchestrator.inc
- 2026-11-07 · Amp
- 2026-11-07 · Google Antigravity
- 2026-11-07 · Augment Code
- 2026-11-07 · Buzz by Block
- 2026-11-07 · Capy
- 2026-11-07 · Claude Code
- 2026-11-07 · Claude Squad
- 2026-11-07 · Cline
- 2026-11-07 · cmux
- 2026-11-07 · CodeLayer / HumanLayer
- 2026-11-07 · OpenAI Codex
- 2026-11-07 · Conductor
- 2026-11-07 · Cursor
- 2026-11-07 · Devin
- 2026-11-07 · Emdash
- 2026-11-07 · Factory
- 2026-11-07 · GitHub Copilot
- 2026-11-07 · Intent by Augment
- 2026-11-07 · Jean
- 2026-11-07 · JetBrains Air
- 2026-11-07 · Jules
- 2026-11-07 · Kiro
- 2026-11-07 · Lanes Desktop
- 2026-11-07 · Linzumi
- 2026-11-07 · Maestro
- 2026-11-07 · Multica
- 2026-11-07 · Omnara
- 2026-11-07 · Omnigent
- 2026-11-07 · Ona
- 2026-11-07 · OpenCode
- 2026-11-07 · OpenHands
- 2026-11-07 · OpenSpec
- 2026-11-07 · Orca
- 2026-11-07 · Paseo
- 2026-11-07 · Podium
- 2026-11-07 · Proliferate
- 2026-11-07 · QM from YC
- 2026-11-07 · Replicas
- 2026-11-07 · Rookery
- 2026-11-07 · Solenta
- 2026-11-07 · super.engineering
- 2026-11-07 · Superconductor
- 2026-11-07 · Paradev
- 2026-11-07 · Synara
- 2026-11-07 · T3 Code
- 2026-11-07 · Tembo
- 2026-11-07 · Unpeel
- 2026-11-07 · Verdent
- 2026-11-07 · Vibe Kanban
- 2026-11-07 · Warp
- 2026-11-07 · Warp Factories / Oz
- 2026-11-07 · Xum
- 2026-11-07 · Zed
- 2026-11-07 · Delta
- 2026-11-07 · Zencoder
- 2026-11-07 · Zenflow Code by Zencoder
- 2026-11-06 · Orchestrator.inc
- 2026-11-06 · Capy
- 2026-11-06 · Podium