AutoAgent is a zero-code framework for creating and deploying LLM agents using natural language. It targets users who want to build sophisticated AI assistants without writing code, offering a user-friendly interface and achieving top performance on benchmarks like GAIA.

How It Works

AutoAgent enables agent and workflow creation through natural language conversations, abstracting away complex coding. It supports a wide range of LLMs and interaction modes (function-calling, ReAct). A key innovation is its "Agentic-RAG" with a self-managing vector database, claimed to outperform existing solutions. The framework is designed to be dynamic, extensible, and lightweight.

Quick Start & Requirements

• Installation: git clone https://github.com/HKUDS/AutoAgent.git, cd AutoAgent, pip install -e .
• Prerequisites: Docker (for containerized environment), API keys for desired LLMs (e.g., OpenAI, Anthropic, Deepseek, Huggingface, Groq).
• Setup: Create a .env file for API keys. CLI mode can be started with auto main or auto deep-research.
• Documentation: https://github.com/HKUDS/AutoAgent (detailed documentation is noted as coming soon).

Highlighted Details

• Top performer on the GAIA benchmark, comparable to OpenAI's Deep Research.
• "Agentic-RAG" with a native self-managing vector database.
• Universal LLM support (OpenAI, Anthropic, Deepseek, vLLM, Grok, Huggingface, etc.).
• Zero-code agent and workflow creation via natural language.

Maintenance & Community

The project was recently renamed from MetaChain to AutoAgent. Community channels include a Slack workspace and a Discord server. Issues can be tracked and discussed on GitHub.

Licensing & Compatibility

The repository does not explicitly state a license in the README. This requires further investigation for commercial use or closed-source integration.

Limitations & Caveats

Tool creation via natural language is temporarily not supported in the workflow editor mode. Detailed documentation is still under development. The project is actively evolving with a roadmap including GUI agents and more benchmark integrations.