riba2534/happyclaw

自托管、多用户、智能体优先的 Agent 工作台

HappyClaw

What it is

HappyClaw is a self‑hosted, multi‑user platform that lets you run Claude Code agents (the "Claude Agent SDK" from Anthropic) as long‑running services. It wraps the official Claude Code runtime in a Node.js/TypeScript server and exposes the agents through a web UI and eight instant‑messaging channels (Feishu, Telegram, QQ, DingTalk, WeChat, Enterprise WeChat, Discord, WhatsApp). Users can create agents, assign them to isolated workspaces, and have those agents execute code, run shell commands, browse the web, or call custom Skills/MCPs.

Core features

Area What you get
Agent‑first workflow Create, edit, version and restore Claude agents; attach custom prompts, Skills, MCPs and plugins.
Workspaces Isolated file system directories; can run on the host machine (admin) or inside a non‑root Docker container (regular users).
Multi‑channel access Unified messaging integration for Feishu, Telegram, QQ, DingTalk, WeChat, Enterprise WeChat, Discord, WhatsApp and a web PWA.
Provider management Configure multiple Anthropic or compatible Claude endpoints; load‑balance (round‑robin, weighted, fail‑over), health‑check and session‑sticky routing.
Scheduling Cron‑style, fixed‑interval or one‑off tasks that run agents or shell scripts, with idempotent run IDs and retryable notifications.
Memory & file handling Workspace‑scoped memory store, versioned file edits, CAS‑style updates, search and terminal access.
Usage & billing Token‑level accounting per agent, workspace and model; CSV export, quotas, subscription and redemption‑code support.
Ops & security RBAC, invitation‑only registration, audit logs, Docker image management, backup/restore, encrypted storage of secrets.
Client experience Real‑time streaming output, tool‑trace visualisation, Markdown/Mermaid/KaTeX rendering, responsive UI and PWA support.

Who might use it

  • Teams that want a private, controllable Claude Code environment instead of a public chat API.
  • Companies needing to run AI‑driven automation (code review, CI assistance, data extraction) that can be triggered from internal chat tools.
  • Developers who want to experiment with Claude agents while keeping execution isolated in Docker containers.
  • Organizations that require fine‑grained permission control, usage tracking and the ability to host the whole stack on‑premises.

Getting started (quick steps)

  1. Clone & start – git clone https://github.com/riba2534/happyclaw.git && cd happyclaw && make start.
  2. Open the web UI at http://localhost:3000 and follow the wizard to create an admin account.
  3. Add an Anthropic (or compatible) Claude provider in Settings → Model & Provider.
  4. (Optional) Connect one or more IM channels under Settings → Message Channels.
  5. Create a Workspace, choose Host (admin) or Container (regular user) execution mode.
  6. Build an Agent (Claude Code) – give it a name, avatar, prompt, and attach any Skills or MCPs you need.
  7. Open a Session (or bind a channel) and start chatting – the agent can read/write files, run shell commands, browse, etc.

Architecture at a glance

Web/PWA  →  HappyClaw Node service (Hono, WebSocket, SQLite)
   │                │
   ├─ Auth/RBAC    ├─ Router & Scheduler
   │                │
   └─ Provider pool (Claude endpoints)
        │
        ├─ Host Runner (direct CLI)
        └─ Docker Agent Runner (isolated containers)
                │
                └─ Claude Agent SDK + Claude Code CLI

The core service handles authentication, routing of messages, task scheduling and persistence. Agent execution is delegated either to the host machine (admin‑approved directories) or to a Docker container that runs the locked‑version Claude Code runtime.

License & community

  • License: MIT (see LICENSE).
  • Stars: ⭐ — the repo shows a badge for GitHub stars.
  • Contributing: Guidelines are in the Contribution section of the README; open issues or pull requests are welcomed.
  • Documentation: API reference, runtime architecture, ACL matrix and deployment guide are provided in the docs/ folder and on the official site https://happyclaw.cc/.

All information above is taken directly from the project's README; no additional features have been inferred.

Related

  • Project
    nanocoai/nanoclaw

    A lightweight AI assistant framework that runs agents in secure, isolated Docker containers and integrates them into various messaging platforms.

  • Project
    composio-community/secure-openclaw

    A personal 24/7 AI assistant that integrates Claude or Opencode with messaging platforms like WhatsApp and Telegram to provide tool access and persistent memory.

  • Project
    johnson7788/MultiUserClaw

    MultiUserClaw is a Docker‑based SaaS framework that lets you run isolated AI assistants (Hermes agents) for many users. It includes a React front‑end, a FastAPI gateway for auth, container management and LLM proxying, per‑user Hermes containers, PostgreSQL storage, and features like knowledge bases, skill store, cron jobs, multi‑channel bots, and multi‑model support.

  • Project
    CoderLuii/HolyClaude

    HolyClaude is a Docker‑based AI development workstation that bundles the official Claude Code CLI, a web UI, headless Chromium, 8 AI‑CLI tools and 50+ dev utilities. It works with any existing Anthropic subscription or API key, supports amd64/arm64, and offers full‑ and slim‑image variants. A quick‑start compose file lets you have a ready‑to‑code environment in under a minute.

  • Project
    fastclaw-ai/fastclaw

    A lightweight AI agent runtime written in Go that allows users to create, manage, and deploy multi-agent systems with sandboxed tool execution and IM channel integrations.