Agor
A self-hosted command center for parallel coding agents, Git branches, and team collaboration.
- Source repo
- preset-io/agor
- Stars
- ★ 1.5k
- Last updated
- today
- License
- NOASSERTION
- Primary language
- TypeScript
- FA score
- 45/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- An engineering lead needs several coding agents to work on separate features or PRs without sharing a working directory or development port.
- Not a fit if
- Teams wanting to commercialize agent orchestration as a third-party product
- Users unwilling to run Node.js 22.12+ and Git
- Source review
- 45/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
Agor is a self-hosted browser workspace for running coding agents such as Claude Code, Codex, and Gemini on isolated Git branches. Each branch is treated as a first-class work unit with its own working directory, development environment, conversation history, and PR. Its architecture combines a FeathersJS daemon, React web UI, oclif CLI, and process-isolated executor, with LibSQL or Postgres storage. The browser session UI exposes structured tool output plus per-prompt token and dollar accounting, while an MCP HTTP endpoint lets agents inspect and operate Agor itself. It can run for an individual developer or support teams with live collaboration, branch-scoped access controls, and progressive Unix isolation.
You start the daemon with the agor CLI, open the web UI, add a repository through onboarding, and create a session for a branch. The daemon in apps/agor-daemon exposes REST, WebSocket, and /mcp?sessionToken=... interfaces; it manages Sessions, Tasks, Messages, Boards, Branches, and Repos, and persists data through Drizzle ORM to LibSQL or Postgres. The process-isolated packages/executor runtime invokes agent SDKs for Claude, Codex, Gemini, OpenCode, and related supported harnesses. Each unit of work receives a Git branch and working directory, while the UI renders conversations and structured tool blocks. Users can Fork a session to copy context, Spawn subsessions with fresh context for focused work, and drop branches into Board Zones to fire templated prompts.
- An engineering lead needs several coding agents to work on separate features or PRs without sharing a working directory or development port.
- A Codex or Claude Code user wants browser-based conversations with visible tool output and per-prompt token and cost accounting.
- A development team wants to organize active Git branches on a visual Board and trigger templated prompts when branches enter a Zone.
- An internal platform team needs a self-hosted agent workspace with branch-scoped ACLs, per-user credentials, and environment variables.
- A team operating long-lived AI coworkers needs searchable knowledge namespaces, skills, and scheduled prompts for each teammate.
How do you install or deploy this agent?
Requires Node.js ≥ 22.12. Run:
npm install -g agor-live
agor init
agor daemon start
agor openagor init creates ~/.agor/ and the database. In the web UI, use the onboarding wizard to add a repository and create the first session. You must provide a usable subscription or credentials for the chosen agent provider; Agor ships no model of its own.
How do you use this agent?
Run agor daemon start, then agor open to enter the web UI. Add a repository in onboarding, create a branch-backed session, and select an available runtime: Claude Code, Codex, Gemini, OpenCode, Copilot, or Cursor (beta). Fork a session to explore an alternative with copied context, or Spawn a subsession for a focused task; placing a branch in a Board Zone can trigger a templated prompt. For source development, run git clone https://github.com/preset-io/agor, cd agor, and docker compose up; then visit http://localhost:5173 and use the documented development login [email protected] / admin.
What are this agent's strengths and limitations?
- Git branches and independent working directories are the core work model, giving parallel agents separate environments, histories, and PRs.
- It supports multiple named coding-agent runtimes per session and does not supply or lock users into its own model.
- Its MCP HTTP endpoint lets agents introspect sessions, branches, and boards and operate the workspace themselves.
- The browser UI provides durable history, structured tool blocks, and per-prompt token and dollar accounting.
- Self-hosted deployment supports LibSQL or Postgres, branch-scoped ACLs, and progressive Unix isolation modes.
- Operation depends on Node.js ≥ 22.12, Git working directories, and the user's own provider subscriptions or credentials.
- Cursor support is explicitly marked beta, so adopting that runtime carries feature-maturity risk.
- The license is BUSL-1.1; before its change date, Agor's agent-orchestration functionality cannot be commercialized as a third-party product or service.
- Multiplayer deployment, strict Unix isolation, Docker, and Postgres are additional configuration paths beyond the minimal quick start.
How does this agent compare with similar options?
Agor does not replace a model provider; it orchestrates existing coding-agent CLIs and SDKs. It explicitly lists Claude Code, Codex, Gemini CLI, GitHub Copilot, OpenCode, and Cursor, adding branch workspaces, browser sessions, an MCP control surface, and multiplayer collaboration around those runtimes.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agor This agent | 45 · Major gaps | CLIFree + model costs | ★ 1.5k | today | TypeScript | Codex · Claude Code |
| Codeg Multi-Agent Coding Workspace | 45 · Major gaps | Desktop appFree + model costs | ★ 4.4k | 2d ago | Rust | Codex · Claude Code · OpenAI API · Claude API |
| Poracode | 55 · Major gaps | Desktop appFree + model costs | ★ 121 | today | TypeScript | ChatGPT · Claude Code |
| Cotal | 77 · Good | CLIFree | ★ 380 | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows RBAC/ACL, Unix isolation modes, per-user credentials and env vars, but no implementation details or audit evidence. User confirmation mechanism not mentioned. Data flow transparency: token and cost accounting present, but data flow not described. Sensitive data handling: credential management mentioned, but encryption or storage not specified. Dependency security: pnpm audit workflow exists, but only critical is blocking, high is advisory. External effects: scheduler and message gateway present, but side-effect control not described. Rollback: not mentioned. Source attribution: license and author info present, but publisher unverified.
Self-consistency: README and architecture diagram consistent, but code implementation not verified. Dependency availability: CI and smoke tests present, but not all dependencies verified. Failure messages: error handling tests present, but no user-visible failure message examples.
Audience and scenarios: targets teams and developers, with docs and quick start. Capability boundaries: supports multiple agent CLIs, but limitations not stated. Trigger precision: scheduler present, but trigger conditions not described. Environment fit: multiple installation methods, but not all platforms covered.
Information architecture: clear docs structure and architecture diagram. Install notes: quick start and extended install. Naming stability: version numbers exist, but naming conventions not stated. Examples and FAQ: examples and docs present, but no FAQ. Known limitations: not explicitly listed. License: BSL 1.1 with clear terms. Versioning and changelog: version numbers present, but no changelog. Maintenance responsibility: community links present, but maintainers not explicit.
Output usability: UI and CLI present, but actual output not verified. Marginal value: unique features, but no comparison with other tools. Cost benefit: token and cost accounting present, but no pricing info.
Claim traceability: README has links, but no specific evidence. Cross-source corroboration: CI and tests present, but not independently verified. Fact-inference separation: not explicitly distinguished.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Publisher identity unverified; treat with caution.
- Dependency audit only blocks critical; high-level vulnerabilities may be ignored.
- User confirmation mechanism absent; high-risk actions may execute automatically.
- Rollback mechanism not mentioned; recovery capability unknown.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.15.0 agor-live v0.15.0
Release notes excerpt
Features - GitHub Copilot SDK integration (beta) — launch and manage Copilot agent sessions with token-level streaming, permission mapping, and MCP support (#811) - Generic Cards & CardTypes system — create custom card types with configurable fields and display them on boards (#812) - MCP SDK migration — migrate internal MCP server to official @modelcontextprotocol/sdk (#816) - Inner tool names for MCP proxy calls — show the actual tool names used inside MCP proxy calls (#835) Fixes - Show MCP OAuth status on session pill and fix browser open race (#836) …View on GitHub ↗ -
v0.14.3 agor-live v0.14.3
Release notes excerpt
Features - Agent SDK slash commands — slash command support with autocomplete for Claude Agent SDK sessions - Session archive/unarchive MCP tools — archive and unarchive sessions via MCP - Board picker search — search filter and recent boards quick-access in board picker - User context for agents — pass user context to agents for multi-user sessions - Required env vars config — block prompts when required user environment variables are missing - Assistant emoji picker — add emoji picker for assistant configuration …View on GitHub ↗ -
v0.14.2 agor-live v0.14.2
Release notes excerpt
Features - Messages MCP tool — add agor_messages_list for browsing and searching session transcripts - AskUserQuestion support — full-stack implementation of interactive agent questions Fixes - Prevent sdk_session_id from being overwritten after first capture - Detect SDK error_during_execution and mark task as failed - Copy-to-clipboard falls back to execCommand when Clipboard API throws - Security: prevent daemon env vars from leaking to agent sessions - Clean up stale zone references when deleting zones …View on GitHub ↗