Agent frameworks for developers: pick a stack
Building an agent is easy to demo and hard to operate. This guide maps the frameworks db.fyi tracks so you can pick a stack without reading every README.
Start here
| Need | Start with | Hub | |------|------------|-----| | General orchestration + tools | LangChain | Agent frameworks | | Document / RAG-heavy apps | LlamaIndex | RAG tools | | Multi-agent crews | CrewAI or AutoGen | Best AI agents | | Visual / low-code flows | Dify, Flowise, n8n | | | Observability | Langfuse, LangSmith | AI for agent observability |
Decision rules
- If retrieval quality is the product, prefer LlamaIndex or Haystack and invest in chunking, embeddings, and evals (AI for RAG, AI for LLM evaluation).
- If tool calling and graphs dominate, LangChain / LangGraph-style stacks or Semantic Kernel fit better.
- If you need many specialized agents, CrewAI or AutoGen; keep a single shared memory and eval harness.
- If non-engineers own the workflow, Dify / Flowise / n8n beat a custom Python service for the first version.
Compare pages
Related tasks
What we will not claim
No framework is “best” in the abstract. Prefer evidence: your latency budget, who maintains the graph, and whether you can evaluate regressions (promptfoo, RAGAS).