BeeAI Framework is a toolkit for building agents and multi-agent systems in Python or TypeScript, hosted by the Linux Foundation under the LF AI & Data program. It covers the pieces an agent system needs: agents, a backend layer that connects to LLM providers through one interface, tools, memory, retrieval-augmented generation, templates, caching, serialization, and workflows for orchestrating several agents.
The Requirement Agent is the framework's approach to predictable behaviour: you attach rules the agent must follow, such as forcing a Think tool at step one, and the agent enforces them across different LLMs. Agents hand work to each other through a HandoffTool, and middleware such as the trajectory logger records tool calls. Agents can be served over protocols including A2A and MCP, and ACP has been folded into A2A under the Linux Foundation.
Features
- Requirement Agent: rule-based constraints that control agent behaviour across LLMs
- Backend: unified interfaces to any LLM provider, including Ollama and watsonx
- Tools: built-in web search, weather, Wikipedia, code execution, and custom tools, plus MCP tools
- RAG: vector stores and document processing for retrieval-augmented generation
- Memory: conversation history with built-in memory strategies
- Templates: dynamic prompts using an extended Mustache syntax
- Serve: host agents in servers that speak A2A and MCP
- Observability: events, logging, and structured error handling with err.explain()
- Cache and serialization: cached calls and saved agent state across sessions
- Workflows: multi-agent execution flows with handoffs between specialised agents