Best AI agent frameworks
Code frameworks for building your own agents with tools, memory and multi-step workflows.
Your tool here
This featured slot is available to tool makers who fit this use case. Clearly labeled, always separate from our picks.
Our picks
1. OpenAI Agents SDK
OpenAI's lightweight SDK for agents with tools, handoffs, guardrails and tracing.
Verdict: Top pick for getting started: small API surface with tools, handoffs and tracing built in.
2. LangGraph
LangChain's framework for stateful, controllable multi-step agent graphs.
Verdict: Best when you need explicit control: state, retries, human-in-the-loop and production reliability.
3. CrewAI
Open-source Python framework for teams of role-based AI agents.
Verdict: Quickest way to prototype role-based multi-agent 'crews'; harder to keep under control at scale.
4. AutoGen
Microsoft's open-source framework for multi-agent conversations and workflows.
Verdict: Microsoft's multi-agent framework; strong for research-style agent conversations.
5. Model Context Protocol
Open standard for connecting AI agents to tools and data.
Verdict: Not a framework, but the standard way to give any agent tools. Build your tools as MCP servers.
Picks are editorial and not paid. Pricing and features change often — check each official site.