riba2534/happyclaw

What it solves

HappyClaw provides a self-hosted, multi-user workbench for Claude Code, allowing users to run AI agents that can interact with project files, execute terminal commands, and use a browser. It solves the problem of accessing powerful agentic capabilities across various platforms by unifying a web interface and eight different instant messaging (IM) channels into a single, persistent service.

How it works

Built on the Claude Agent SDK for TypeScript, HappyClaw encapsulates the Claude Code runtime into a service. It uses a hierarchical model: Agents (identity and skills) $\rightarrow$ Workspaces (file and execution boundaries) $\rightarrow$ Runtime Sessions (conversation contexts).

Execution happens in two modes:

• Host Mode: Runs directly on the host machine in authorized directories (for admins).
• Container Mode: Runs in isolated non-root Docker containers (for general members).

It integrates with multiple Claude providers (official and third-party) and supports a wide array of IM channels including Feishu, Telegram, QQ, DingTalk, WeChat, Enterprise WeChat, Discord, and WhatsApp via a Node.js core service.

Who it’s for

• Developers and teams who want to self-host their AI coding agents.
• Users who need to interact with their AI agents via IM apps rather than just a terminal or web UI.
• Administrators requiring secure, isolated execution environments for multiple users.

Highlights

• Multi-Channel Access: Unified access via Web, PWA, and 8 major IM platforms.
• Agent-First Architecture: Clear separation between agent identity, workspace files, and session history.
• Hybrid Execution: Supports both native host execution and Docker sandbox isolation.
• Extensible Capabilities: Integration with MCP (Model Context Protocol), custom Skills, and Claude Code Plugins.
• Provisioning & Ops: Includes built-in RBAC, usage statistics, billing/quota management, and cron-based scheduled tasks.