metabot  by xvirobotics

Build supervised, self-improving AI agent organizations

Created 1 month ago
523 stars

Top 60.2% on SourcePulse

GitHubView on GitHub
Project Summary

This project provides infrastructure for building supervised, self-improving agent organizations, enabling users to run sophisticated AI agents like Claude Code with shared memory, task scheduling, and inter-agent communication accessible via Feishu and Telegram. It aims to empower companies to operate as "agent-native" organizations, managed by a small human team overseeing AI agents.

How It Works

MetaBot orchestrates AI agents using a Claude Code Kernel for each. MetaSkill acts as an agent factory, capable of generating specialized agent teams. MetaMemory provides a persistent, shared knowledge store with full-text search and a web UI, allowing agents to learn and share information across sessions without retraining, and can sync to Feishu Wiki. The IM Bridge integrates with Feishu and Telegram for human supervision and interaction, while an Agent Bus (REST API) facilitates inter-agent communication, task delegation, and bot management. A Task Scheduler handles autonomous, recurring, or delayed tasks, creating a robust system for self-improving agent organizations.

Quick Start & Requirements

  • Primary install / run command: A one-line bash or PowerShell script for Linux/macOS (curl -fsSL https://raw.githubusercontent.com/xvirobotics/metabot/main/install.sh | bash) or Windows (irm https://raw.githubusercontent.com/xvirobotics/metabot/main/install.ps1 | iex). Manual installation involves git clone, npm install, and configuration.
  • Non-default prerequisites and dependencies: Node.js 20+, Claude Code CLI installed and authenticated. Requires configuration of Feishu or Telegram bot tokens and app permissions.
  • Links:
    • Linux/macOS Installer: https://raw.githubusercontent.com/xvirobotics/metabot/main/install.sh
    • Windows Installer: https://raw.githubusercontent.com/xvirobotics/metabot/main/install.ps1
    • Claude Code CLI: https://github.com/anthropics/claude-code
    • Feishu App Setup: open.feishu.cn

Highlighted Details

  • Agent Factory (MetaSkill): Automates the creation of complete agent teams (orchestrator, specialists, reviewers) for specific projects.
  • Persistent Shared Memory (MetaMemory): An embedded SQLite knowledge store with full-text search, a web UI (http://localhost:8100), and optional auto-sync to Feishu Wiki.
  • Cross-Platform Chat (IM Bridge): Enables interaction with agents via Feishu/Lark and Telegram, including mobile access.
  • Autonomous Task Scheduling: Supports cron-like scheduling for recurring tasks and one-time delays, with persistence across restarts.
  • Agent Communication Bus: A REST API for inter-agent task delegation, bot CRUD operations, and scheduling.
  • Bypass Permissions Mode: Claude Code operates autonomously, requiring careful access control via IM platforms and cost capping.

Maintenance & Community

MetaBot is developed by XVI Robotics, who utilize it internally to run their company as an "agent-native organization." The project is open-sourced with the stated vision of shaping future company operations. Specific community channels like Discord or Slack are not detailed in the README.

Licensing & Compatibility

  • License type: MIT License.
  • Compatibility notes: The MIT license is generally permissive for commercial use. Supports any Anthropic-compatible API, including Kimi/Moonshot, DeepSeek, and GLM/Zhipu.

Limitations & Caveats

Claude Code runs in bypassPermissions mode, granting full read/write/execute access to the working directory, necessitating strict access control via IM platforms and cost capping (maxBudgetUsd). Agent communication is currently synchronous request-response, with asynchronous bidirectional protocols planned for future development.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
101
Issues (30d)
29
Star History
420 stars in the last 30 days

Explore Similar Projects

Starred by Andrew Ng Andrew Ng(Founder of DeepLearning.AI; Cofounder of Coursera; Professor at Stanford), Jack Lukic Jack Lukic(Author of Semantic UI), and
5 more.

ag2 by ag2ai

0.5%
4k
AgentOS for building AI agents and facilitating multi-agent cooperation
Created 1 year ago
Updated 21 hours ago
Feedback? Help us improve.