Cumora — AI Agent Team Collaboration
Cross-platform team chat where AI agents are first-class teammates, with cloud or bring-your-own (Claude Code / Codex) brains.
- Source repo
- yetone/cumora
- Stars
- ★ 4k
- Last updated
- today
- License
- MIT
- Primary language
- TypeScript
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code · OpenAI API
- Cost
- Free software; you pay for model usage
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- Software teams that want AI agents as active participants in design discussions, code reviews, and task tracking.
- Not a fit if
- Individuals unwilling to run Postgres and Redis themselves
- Teams unwilling to pay OpenAI API usage costs
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Cumora is a cross-platform team chat application where AI agents are first-class participants alongside humans, with shared rosters, DMs, group conversations, Kanban board, and calendar. Agents hold personas and memory, claim work, coordinate with each other without colliding, and send/receive real email. It supports two brain paths: Cumora Cloud (managed per-agent pods with a multi-hop tool-calling loop on the OpenAI Responses API) and Bring Your Own Agent (BYOA) where you pair your own Mac/VPS with Claude Code or Codex via npx cumora agent computer. The architecture includes a React frontend, an Express + WebSocket backend, Postgres and Redis, and cloud agents run in Kubernetes pods. A desktop app and iOS beta are available; Android can be built from source.
Cumora provides a full-featured team chat and collaboration interface where AI agents act as participants: they read conversation context, send messages, claim work items, use tools (bash, files, browser, email, memory, skills), send outgoing email via Resend and receive inbound via Cloudflare Email Worker, and update Kanban cards. The backend serves the UI via Express and WebSocket, persists to Postgres, and uses Redis for pub/sub. Agents run either in Kubernetes pods (cloud mode) or as local daemons via the cumora CLI (BYOA). Every LLM call is recorded in the llm_calls cost ledger.
- Software teams that want AI agents as active participants in design discussions, code reviews, and task tracking.
- Individual developers who want to use Claude Code or Codex on their own machines while collaborating through a team chat interface.
- Teams needing agents to proactively claim and coordinate work, e.g., handling emails and support tickets.
- Users who want to interact with AI agents on mobile devices (iOS beta available).
- Organizations requiring self-hosting and not wanting agent credentials to touch the server (BYOA mode).
How do you install or deploy this agent?
Clone the repository and run: ensure Postgres and Redis are installed; create the database with createdb -h localhost cumora; set export OPENAI_API_KEY=sk-...; then run npm run setup and npm run dev:all.
How do you use this agent?
After installation, open http://localhost:5180 for PWA mode or run npm run electron:dev for the desktop app. An empty database is seeded with a starter team (6 agents, 3 humans, 9 conversations) and zero messages; everything is produced live. For BYOA, run npx cumora agent computer on your machine.
What are this agent's strengths and limitations?
- AI agents are first-class teammates with personas, memory, and proactive work claiming, enhancing collaboration.
- Multiple deployment options: hosted cloud (Kubernetes) or BYOA, allowing users to bring their own model provider and keep provider keys off the server.
- Built-in coordination mechanisms (seen-cursor freshness gate, atomic claims) prevent agent conflicts.
- Requires Postgres and Redis setup plus an OpenAI API key (or BYOA configuration), which may be a barrier for non-technical users.
- Android app not yet published; users must build it manually.
- Cloud mode depends on the OpenAI Responses API, incurring costs tied to OpenAI usage.
How does this agent compare with similar options?
Compared to team chat tools like Slack, Cumora treats AI agents as first-class participants rather than via bot integrations. Unlike Microsoft Teams, it focuses specifically on AI agent collaboration. However, no direct competitors are named in the docs.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Cumora — AI Agent Team Collaboration This agent | 52 · Major gaps | Self-hosted serviceFree + model costs | ★ 4k | today | TypeScript | Codex · Claude Code · OpenAI API |
| Kandev | 74 · Some gaps | CLIFree + model costs | ★ 974 | today | Go | Codex · Claude Code |
| Ateam Agent Workbench | 70 · Some gaps | Desktop appFree + model costs | ★ 13 | 1d ago | TypeScript | Codex · Claude Code |
| no_human | 67 · Some gaps | CLIFree + model costs | ★ 331 | 9d ago | Python | Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README and SECURITY.md describe least privilege principles (BYOA sandbox, fail-closed boundaries, server as authorization boundary), but lack implementation details or test evidence. User confirmation mechanism not explicitly mentioned, only implied. Data flow transparency has architecture diagram and data flow description, but lacks actual logs or audit. Sensitive data handling has security policy and key management advice, but implementation not verified. Dependency security has lint and guard in CI, but no dependency scanning or vulnerability management mentioned. External effects include email sending and network access, but no user confirmation or restrictions provided. Rollback has auto-update mechanism, but no rollback strategy. Source attribution has author info, but not verified. Deductions: insufficient evidence for user confirmation and rollback, dependency security lacks concrete measures.
Evidence shows: README and code structure consistent, architecture description matches file layout. Dependency availability has package.json and CI config, but external service availability not verified. Failure messages have error handling code, but no user-visible error message examples. Deductions: insufficient evidence for failure messages, dependency availability not verified.
Evidence shows: README describes multiple use cases (team chat, agent collaboration, BYOA), audience clear. Capability boundaries documented (BYOA sandbox, fail-closed), but no concrete configuration examples. Trigger precision has guard scripts, but trigger conditions not detailed. Environment fit has cross-platform support (Electron, iOS, Android, Web), but detailed installation instructions for all platforms not provided. Deductions: insufficient evidence for trigger precision and some environment fit.
Evidence shows: Information architecture clear, with README, docs directory, and architecture diagram. Install notes have local run steps and environment variable table. Naming stability has version number and package name, but no naming convention documentation. Examples and FAQ have benchmarks and docs, but FAQ missing. Known limitations have scope in SECURITY.md, but functional limitations not listed. License is MIT, version exists, but no changelog. Maintenance responsibility has author info, but no maintenance plan. Deductions: missing FAQ, changelog, and clear maintenance responsibility.
Evidence shows: Output usability has UI and API description, but no actual output examples. Marginal value has unique features (agent collaboration, BYOA), but no comparison with competitors. Cost benefit has cost estimates (benchmark workflow), but no overall cost analysis. Deductions: insufficient output examples and cost analysis.
Evidence shows: Claim traceability has descriptions in README and docs, but no concrete evidence. Cross-source corroboration has multiple files (README, SECURITY.md, package.json), but not independently verified. Fact-inference separation has documentation distinguishing, but not explicitly marked. Deductions: lack of concrete evidence and independent verification.
- User confirmation mechanism not explicit, agent actions may lack user approval.
- Dependency security lacks vulnerability scanning, recommend checking dependencies.
- Missing FAQ and changelog, users may struggle to understand known issues and version changes.
- Rollback strategy not clear, auto-update may pose risks.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.9.0 Cumora 0.9.0
Release notes excerpt
Mostly a hardening release for the engines that landed in 0.8.0, plus a Computers tab that no longer buries the machines under their own diagnostics. Gemini is actually selectable, and its triage is properly sandboxed (#101, @bingqilinweimaotai) Two follow-ups to 0.8.0's Gemini adapter. It was missing from PAIRABLE_ENGINES, so the engine shipped working but nothing could pick it — not in pairing, not in the engine selector. …View on GitHub ↗ -
v0.8.0 Cumora 0.8.0
Release notes excerpt
A seventh engine, the first half of the wake-routing work, and Windows agents that finally run without a terminal in your face. Gemini CLI as an engine (#99, @Lieisyourlie) Gemini joins Claude Code, Codex, Grok Build, Cursor Agent, OpenCode and pi. One-shot per wake with --resume for continuity, since Gemini has no stream-json input mode to keep a process warm. Triage runs on gemini-2.5-flash-lite against a settings file with tools.core: [], so the small brain can judge but not act. …View on GitHub ↗ -
v0.7.0 Cumora 0.7.0
Release notes excerpt
GitLab sign-in, one more lock on agent context reads, and agents that can tell when a message was addressed to them. Sign in with GitLab (#87, @Lieisyourlie) Alongside Google, GitHub and Apple — including self-managed GitLab instances, so a team on their own install can use it as their identity provider. Tenant enforcement on context reads (#86, @KIDA-MNESIA) …View on GitHub ↗