The four open-source frameworks that make building AI agents practical. Each takes a different approach — different architecture, different complexity, different strengths. This hub covers all four with complete guides.
Stuck at any point? Ask an AI about this page →The four frameworks
LangChain is the largest ecosystem for building with LLMs — 1,000+ integrations, massive documentation, broad community. LangGraph (LangChain's own recommendation for agents) adds a graph-based state machine on top, making complex multi-step and multi-agent workflows reliable and inspectable.
Full guide → Easiest to startRole-based multi-agent framework where agents are defined with a role, goal, and backstory — just like hiring a team. The most intuitive way to build multi-agent systems. Agents collaborate on tasks sequentially or in parallel, with an orchestrator managing the crew.
Full guide → Maintenance modeConversation-based multi-agent framework from Microsoft Research. Agents communicate by sending messages to each other. Pioneered many patterns now standard across frameworks. Important caveat: AutoGen is being merged into Microsoft's Agent Framework — no new features, bug fixes only.
Full guide → Best for dataSpecialises in data-augmented agents — AI agents that reason over large document sets, knowledge bases, and structured data using RAG (Retrieval-Augmented Generation). The go-to choice when your agent needs to work with proprietary documents, internal data, or large knowledge bases.
Full guide →Side-by-side comparison
| Feature | LangChain/LangGraph | CrewAI | AutoGen | LlamaIndex |
|---|---|---|---|---|
| Language | Python + JS/TS | Python | Python | Python + TS |
| Primary pattern | Graph state machine | Role-based crew | Conversational | RAG + agents |
| Multi-agent | Yes | Yes (core feature) | Yes | Yes |
| Learning curve | Steep | Gentle | Moderate | Moderate |
| Best for | Complex pipelines | Getting started fast | Research / legacy | Document-heavy tasks |
| Active development | Yes | Yes | No — maintenance only | Yes |
| MCP support | Yes (LangGraph) | Yes | Partial | Yes |
| Licence | MIT | MIT | MIT (CC-BY-NC for some) | MIT |
Understand agentic AI first
Opens your assistant with this page as the source, and a question rather than a summary. It will ask what you are building before it answers.
Nothing is sent from here. The link carries only this page’s title and address.