AgentHeavenยถ
Ask not what your agents can do for you, ask what you can do for your agents.
AgentHeaven is NOT another agent framework; itโs an approach to knowledge management designed for the era of AI.
The purpose of AgentHeaven is to provide everything BEFORE the agent acts, enabling agents to work in a friendly, context-rich environment, i.e., a โheavenโ for agents.
It empowers you to build a agentic lifelong learning system for any data application, treating AI agents as capable human beings, and as our users.
๐ English Documentation ๐ ไธญๆๆๆกฃ ๐ป Documentation GitHub
Key Featuresยถ
๐ Knowledge Management: Turn domain knowledge from docs, databases, and user queries into Unified Knowledge Format (UKF) and manage them efficiently and automatically.
๐๏ธ Databases Integration: Highly-abstract integration with SQL databases (via SQLAlchemy), vector databases (via LlamaIndex), and other storage & retrieval backends.
๐ฆ LLMs Integration: Unified interface for various language models (via LiteLLM) with configurable presets and modular caching.
๐ Imitation is All You Need: An agent builder based on imitation that creates and continuously refines domain-specific agents through weak-supervision.
โก Agents For Agents: Use agents to help improve agents, starting the exponential growth of agentic system development.
๐ค Human-In-The-Loop: Get involved in the agentsโ training process easily using natural language instructions.
๐ Internationalization: Using Jinja + Babel for prompt management, with agentic translation and prompt creation features.
๐ Utilities: A comprehensive collection of Python helper functions to speed up the development process.
๐ฅ CLI & GUI Tools: Command-line and graphical interfaces integrated, for both developers and low-code users.
๐ MCP Compatible: Centered around FastMCP 2.0 to support various tool use and function call interfaces.