countbot-ai/CountBot

What it solves

CountBot provides a lightweight AI Agent framework and central hub designed specifically for Chinese users. It solves the problem of integrating disparate LLMs, messaging channels, and external tools into a single, maintainable system that can run long-term in local or private environments.

How it works

It acts as a central hub connecting four key layers:

• Models: Connects to 100+ LLM providers with customizable model strategies.
• Channels: Integrates with various IM platforms including WeChat, Feishu, DingTalk, Telegram, and QQ.
• Organization: Manages roles, teams, workflows, and memory boundaries.
• Execution: Connects to local files, Shell, Web, and external industry tools like Claude Code, Codex, and OpenCode.

Who it’s for

• Users who want to build complex tasks using multi-agent collaboration (roles collaborating to complete a goal).
• Individuals or teams needing a locally hosted AI assistant or automation flow that operates across multiple messaging apps.
• Developers looking for an open-source infrastructure to integrate tool calling, file operations, and scheduled tasks into one environment.

Highlights

• Multi-Agent Collaboration: Supports pipeline, graph, and council coordination modes.
• Omni-Channel Support: A single workspace can serve multiple bots across different IM platforms.
• Extensible Tooling: Includes built-in tools for Shell, Web, and screenshots, plus MCP (Model Context Protocol) client support for expanded execution capabilities.
• Long-term Operation: Features Cron jobs, heartbeat mechanisms, and session memory (summaries and long-term storage) for persistent autonomy.
• Knowledge Integration: Includes a Wiki knowledge base with BM25 full-text search and integration with IMA knowledge bases/notes.