A token-efficient, microkernel AI agent for CLI, Web UI and chat channels that routes each turn to the cheapest capable model.
poco-ai/poco-claw
A self-hosted, Web UI-based personal AI coworker powered by a Claude Code agent, with IM integration and sandboxed task execution.
Open-source alternative to:OpenClaw
- Stars
- 1,357
- Forks
- 130
- Open issues
- 3
- Last commit
- 3 days ago
What is it?
Poco is an open-source alternative to OpenClaw that wraps a Claude Code-based agent in a Web UI. Users give the agent tasks from a browser, mobile device, or chat apps such as DingTalk, Feishu, and Telegram, and each task runs inside an isolated Docker container so installs, file edits, and commands do not touch the host system.
The project is organized around long-lived servers with channels, DMs, and reply threads. Persistent agents can be mentioned in conversations, messages can be turned into trackable tasks, and outputs can be published to shared files. Projects hold defaults for model, presets, Git repos, local directory mounts, and project files. Local directory mounting is available only in self-hosted deployments.
The agent supports Plan Mode, Slash Commands, MCP servers, Skills, a built-in browser, and scheduled background runs. An artifacts viewer previews formats such as HTML, PDF, Markdown, and Excalidraw, and mem0 provides persistent memory. Deployment is handled by a quickstart script and Docker.
Key features
- 01Channel-based team collaboration with persistent agents that can be mentioned in conversations
- 02Sandboxed execution in isolated Docker containers
- 03Artifacts preview for HTML, PDF, Markdown, images, and diagram formats
- 04IM integration with DingTalk, Feishu, and Telegram
- 05MCP and Skills support, plus a built-in browser for web research
- 06Long-term memory powered by mem0
Use cases
- Running a shared AI assistant for a small team that works through channels and shared files
- Delegating coding or research tasks to an agent while away from the desk via mobile or chat apps
- Self-hosting an agent that works on local project directories inside a container
Topics
FAQ
- What is poco-claw?
- A self-hosted, Web UI-based personal AI coworker powered by a Claude Code agent, with IM integration and sandboxed task execution. Poco is an open-source alternative to OpenClaw that wraps a Claude Code-based agent in a Web UI. Users give the agent tasks from a browser, mobile device, or chat apps such as DingTalk, Feishu, and Telegram, and each task runs inside an isolated Docker container so installs, file edits, and commands do not touch the host system.
- Is poco-claw free to use? What license does it have?
- poco-claw is open source under the MIT license. Check the license terms before using it in commercial products.
- Is poco-claw still maintained?
- Actively maintained. The last commit to poco-claw was on Oct 19, 2026.
- What can I use poco-claw for?
- Running a shared AI assistant for a small team that works through channels and shared files Delegating coding or research tasks to an agent while away from the desk via mobile or chat apps Self-hosting an agent that works on local project directories inside a container
- Where can I get poco-claw?
- The source code and installation instructions are on GitHub: https://github.com/poco-ai/poco-claw
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Open-source multi-agent platform that gives each AI agent its own cloud workspace with filesystem, desktop, browser, network and long-term memory.
A self-hosted personal AI assistant that runs on your machine or in the cloud and connects to chat apps like DingTalk, Lark, Discord and Telegram.
Self-hosted Java/Spring Boot AI assistant platform with multi-user workspaces, multi-provider model failover, an LLM Wiki, MCP and Skills, and IM channel integrations.
LobeHub is a platform for organizing AI agents into always-on teams that can be hired, scheduled, and reviewed from one place.
A private, local-first personal AI agent runtime that runs sandboxed tools, keeps memory and knowledge on your machine, and works with local, cloud, or Grok models.