OpenCrew Team OS
Run a role-based OpenClaw AI team from the chat workspace your team already uses.
- Source repo
- AlexAnys/opencrew
- Stars
- ★ 496
- Last updated
- 7mo ago
- License
- MIT
- Primary language
- Shell
- FA score
- 48/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specific
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An existing OpenClaw user who wants separate Slack channels for executive intent, technical design, and implementation handled by CoS, CTO, and Builder roles.
- Not a fit if
- Teams wanting a standalone multi-agent runtime without an existing OpenClaw install
- Teams needing Discussion-mode multi-agent collaboration on Feishu or Discord
- Teams unwilling to pay higher total token usage than a single-agent setup
- Source review
- 48/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
OpenCrew is a multi-agent collaboration system built on OpenClaw for decision makers who want to operate an AI team through Slack, Feishu, or Discord. It maps channels or groups to roles and threads to tasks, with workspaces for roles such as CoS, CTO, Builder, CIO, Research, KO, and Ops. Its operating model includes L0–L3 autonomy levels, QAPS task types, structured Closeouts and Checkpoints, knowledge extraction by KO, and change auditing by Ops. Agents can delegate work through sessions_send; on Slack, a separately configured Slack App enables @mention-based Discussion collaboration. It is deployed into an existing OpenClaw installation with connected chat platforms, rather than offered as a standalone hosted service.
During deployment, OpenCrew adds role workspaces and the shared protocols under shared/ to an OpenClaw setup, then configures channel-to-agent routing in openclaw.json. Users start work in channels such as #hq, #cto, and #build, or their equivalent groups, while each agent responds within its assigned domain and organizes tasks in threads. Delegation uses a visible channel anchor plus sessions_send as a two-step handoff. In Slack Discussion mode, an orchestrating agent with an independent Slack App joins an execution channel and collaborates through @mentions in the same thread. Work can produce Closeouts, Checkpoints, and reusable knowledge distilled by KO; system changes are subject to an Ops review loop.
- An existing OpenClaw user who wants separate Slack channels for executive intent, technical design, and implementation handled by CoS, CTO, and Builder roles.
- A decision maker managing several concurrent initiatives who wants channels to show responsibilities and threads to keep each task separate.
- A Slack team that needs a coordinator and an execution agent to discuss, review, and reach agreement in a shared task thread.
- A user who wants project learnings compressed into structured Closeouts and later distilled into reusable operating knowledge by KO.
- An OpenClaw user on Feishu or Discord who wants the same role-per-channel operating model while accepting platform-specific limitations.
How do you install or deploy this agent?
Prerequisite: an existing OpenClaw installation for which openclaw status succeeds. Connect Slack, Feishu, or Discord; create #hq, #cto, and #build (or equivalent groups) and invite the bot. Clone the repository with git clone https://github.com/AlexAnys/opencrew.git /tmp/opencrew. Then have the existing OpenClaw follow the repository’s DEPLOY.md incremental deployment flow: provide platform credentials (Slack Bot Token and App Token; Feishu App ID and App Secret; or Discord Bot Token), map channels to CoS/CTO/Builder, and preserve existing models, auth, and gateway settings.
How do you use this agent?
After deployment, send a message in each CoS, CTO, and Builder channel or group to confirm that its assigned agent replies. Ask the CTO to delegate a task to Builder and verify that a response appears in Builder’s channel. For one-way handoffs, configure A2A Delegation with agentToAgent.allow and maxPingPongTurns, then add the Delegation rules to the participating roles. For Slack discussion collaboration, create an independent Slack App for at least one key agent, configure multiple accounts while retaining accounts.default, invite the new bot into the target channel, and verify that both agents reply to @mentions in one thread.
What are this agent's strengths and limitations?
- Uses a concrete channel/group-to-role and thread-to-task model, making team responsibilities and work visible in the chat workspace.
- Offers both Delegation and Slack Discussion A2A modes; the latter supports review and discussion by independently identified bots in execution channels.
- Combines structured Closeouts, Checkpoints, three-layer knowledge capture, and Ops Review in one operating model.
- Supports Slack, Feishu, and Discord, with a documented three-role starting configuration of CoS, CTO, and Builder.
- Core operation depends on OpenClaw, so it is not a standalone general-purpose multi-agent runtime.
- The documented, validated Discussion mode is Slack-specific and requires an additional Slack App plus multi-account configuration.
- A2A @mention checks, turn counting, and
NO_REPLYdiscipline are prompt-enforced; the repository recommends low-risk testing for models other than Claude Opus 4.6. - Feishu lacks thread task isolation, and Discord Discussion mode is blocked by an OpenClaw code-level bug.
- The model uses more total tokens than a single-agent setup, and Slack’s free tier retains message history for only 90 days.
How does this agent compare with similar options?
The repository distinguishes OpenCrew from developer SDKs such as CrewAI and AutoGen: those are presented as code-based agent orchestration tools, while OpenCrew is positioned around managing an AI team through Slack with explicit roles, tasks, knowledge capture, and governance.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OpenCrew Team OS This agent | 48 · Major gaps | CLIFree + model costs | ★ 496 | 7mo ago | Shell | — |
| Switch AI Team Workspace | 41 · Major gaps | Desktop app | ★ 3.9k | today | Python | — |
| dot-skill — Digital Life Distiller | 49 · Major gaps | Agent plugin / skillFree + model costs | ★ 26k | 24d ago | Python | Codex · Claude Code · OpenAI API |
| MateClaw | 56 · Major gaps | Self-hosted serviceFree + model costs | ★ 1.2k | 1d ago | Java | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README describes Autonomy Ladder (L0-L3), with L3 requiring user confirmation and L2 reporting after action, indicating user confirmation mechanism; deployment prompts explicitly instruct not to modify models/auth/gateway config, only add OpenCrew increments, reflecting least privilege; but data flow transparency is insufficient, not detailing how sensitive data (e.g., Slack tokens) is stored and accessed; dependency security not mentioned, no dependency list or security audit; external effects described (e.g., Slack message sending) but not limited; rollback mechanism only mentioned in A2A setup guide, not detailed; source attribution only via MIT license and contributor statement, not verified. Deductions: missing dependency security, insufficient data flow transparency, unclear rollback.
Evidence shows: README and documentation structure are consistent, describing stable and experimental features, good self-consistency; dependency availability not explicit, no OpenClaw version requirements or platform compatibility details; failure messages mentioned (e.g., common errors, troubleshooting checklist) but not detailed. Deductions: insufficient dependency availability info, failure messages not specific.
Evidence shows: target audience clear (decision-makers, non-technical users), scenarios cover Slack, Feishu, Discord with platform-specific guides; capability boundaries via stable/experimental features, but not each agent's specific permission boundaries; trigger precision via channel binding and @mention, but not detailed; environment fit considers platforms, but Feishu thread support limited. Deductions: capability boundaries and trigger precision not detailed.
Evidence shows: clear information architecture with directory structure and doc guides; detailed install notes with automated and manual steps; naming stability not explicit but agent names consistent; examples and FAQ rich; known limitations documented; MIT license clear; versioning changelog absent but status updates present; maintenance responsibility not explicit but PRs welcome. Deductions: missing versioning changelog, unclear maintenance responsibility.
Evidence shows: output usability via Closeout mechanism and structured outputs, but no actual output examples; marginal value clear, solving context bloat etc.; cost-benefit discussed (token usage) but no concrete data. Deductions: missing output examples, insufficient cost-benefit data.
Evidence shows: claims in README (e.g., 'tested stable') lack test evidence; cross-source corroboration insufficient, relying on single repo; facts and inferences not clearly separated, e.g., 'A2A v2 is the best way' is subjective. Deductions: claims lack traceability, facts and inferences mixed.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Dependency security not mentioned; check OpenClaw and its dependencies for known vulnerabilities.
- Sensitive data handling (e.g., Slack tokens) storage and access paths are opaque; be cautious.
- Rollback mechanism only mentioned in A2A setup guide, not detailed; confirm before deployment.
- Versioning changelog missing; cannot track version evolution.
- Claims (e.g., 'tested stable') lack test evidence; verify independently.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.3.0 v0.3.0 — A2A v2: Agent 间真正的协作
Release notes excerpt
A2A v2：找到了 Agent 间协作的最佳方式 这是 OpenCrew 自诞生以来最重大的架构更新。Agent 之间不再只能单向派任务——现在可以在同一个 Slack 频道里像同事一样讨论、评审、迭代。 核心变化 之前（v0.2.x）：所有 Agent 共享一个 Slack bot → bot 不能触发自己 → Agent 间只能靠 sessions_send 单向委派。 现在（v0.3.0）：给至少一个关键 Agent 创建独立 Slack App → 拉进执行 Agent 的频道 → 频道内高维度对话 + 基于文件的实际协作 + 用户 review 最终产出。 <p> <img src="https://raw.githubusercontent.com/AlexAnys/opencrew/main/docs/OpenCrew-A2A-V2.svg" width="48%" alt="A2A v2 架构"> <img src="https://raw.githubusercontent.com/AlexAnys/opencrew/main/assets/screenshots/a2a-v2-discussion-demo.png" width="48%" alt="A2A v2 实战"> </p> …View on GitHub ↗ -
v0.2.2 v0.2.2 — 独立 Bot 身份：飞书多应用 + Discord 三方案
Release notes excerpt
🚀 v0.2.2 — 独立 Bot 身份：飞书多应用 + Discord 三方案 在 [v0.2.1](https://github.com/AlexAnys/opencrew/releases/tag/v0.2.1) 的三平台支持基础上，本版本为飞书和 Discord 新增了独立 Bot 身份方案 — 让每个 Agent 拥有自己的名字和头像。 背景 v0.2.1 中三个平台统一使用"单 Bot 模式"：一个 bot 加入多个频道，通过频道路由到不同 Agent。这是最简配置，但所有 Agent 共享同一外观。 飞书和 Discord（不同于 Slack）天然支持多 Bot 共存。本版本将这一能力文档化，为需要更好用户体验的团队提供进阶选择。 ✨ 新增 - 飞书 — 多应用模式（[进阶章节](docs/FEISHU_SETUP.md)） - 为每个 Agent 创建独立飞书自建应用，获得独立名称、头像和 API 配额 - 利用 OpenClaw 原生 accounts 多账户配置，无需 hack - 适用场景：正式生产环境、需要权限隔离或独立限速 - Discord — Bot 身份方案选择（[进阶章节](docs/DISCORD_SETUP.md)） …View on GitHub ↗ -
v0.2.1 v0.2.1 — 多平台支持：飞书 + Discord
Release notes excerpt
🚀 v0.2.1 — 多平台支持：飞书 + Discord OpenCrew 现在支持 Slack · 飞书 · Discord 三个平台。选择你熟悉的平台作为指挥中心，部署流程和配置参考均已覆盖。 ✨ 新增 - [飞书接入指南](docs/FEISHU_SETUP.md)（中文 + [English](docs/en/FEISHU_SETUP.md)） - 从创建飞书自建应用到连接 OpenClaw 的完整步骤 - WebSocket 长连接模式，无需公网服务器 - ⚠️ 明确标注 Thread 限制（OpenClaw 飞书插件暂不支持话题隔离） - Lark（国际版）接入说明 - [Discord 接入指南](docs/DISCORD_SETUP.md)（中文 + [English](docs/en/DISCORD_SETUP.md)） - 从创建 Discord Bot 到连接 OpenClaw 的完整步骤 - Gateway Intents 配置说明（含 Message Content Intent） - Thread 继承机制说明 …View on GitHub ↗