The best AI agent frameworks
If you are building your own agent, a framework saves you from writing tool calling, memory and orchestration from scratch. Every framework in the AgentDrawer directory is open source; they differ in language, in how they model multi-agent work and in what they emphasize, from data retrieval to long-term memory.
Quick comparison
| Tool | Type | Pricing | In one line |
|---|---|---|---|
| LangChain | Python / JS | Open Source | Framework (and LangGraph) for building LLM agents. |
| CrewAI | Python | Open Source | Framework for orchestrating role-playing multi-agent crews. |
| AutoGen | Python / .NET | Open Source | Microsoft's framework for multi-agent applications. |
| LlamaIndex | Python / TS | Open Source | Framework for agents over your data. |
| OpenAI Agents SDK | Python / TS | Open Source | Lightweight SDK for building multi-agent workflows. |
| Pydantic AI | Python | Open Source | Type-safe Python agent framework from the Pydantic team. |
| Mastra | TypeScript | Open Source | TypeScript framework for AI agents and workflows. |
| Letta | Python / REST API | Open Source | Platform for stateful agents with long-term memory. |
How to choose
For TypeScript projects, Mastra is built for the language. In Python, Pydantic AI and the OpenAI Agents SDK are lightweight and quick to learn, while LangChain with LangGraph is the most widely adopted and the most flexible for complex, stateful flows. CrewAI and AutoGen focus on teams of collaborating agents. LlamaIndex is strongest when the agent mainly works over your own documents, and Letta when agents need persistent long-term memory.
Every option, in detail
1. LangChain
LangChain is one of the most widely used open-source frameworks for building LLM applications and agents, alongside LangGraph for stateful agent workflows. The company now describes it as an open agent engineering platform. The frameworks are free; LangSmith is its paid platform.
Pricing: Open Source · Website (opens in new tab)
2. CrewAI
CrewAI is an open-source Python framework for orchestrating teams of role-playing AI agents that collaborate on tasks. It also offers a commercial multi-agent platform for deploying and managing crews in production.
Pricing: Open Source · Website (opens in new tab)
3. AutoGen
AutoGen is Microsoft's open-source framework for building multi-agent AI applications. Agents can converse with each other, use tools and work with humans in the loop, and it includes AutoGen Studio for low-code prototyping.
Pricing: Open Source · Website (opens in new tab)
4. LlamaIndex
LlamaIndex is an open-source framework for building agents and workflows over your own data. The company also offers LlamaParse and related cloud tools for document OCR and parsing. The core framework is free to use.
Pricing: Open Source · Website (opens in new tab)
5. OpenAI Agents SDK
The OpenAI Agents SDK is a lightweight open-source library for building multi-agent workflows. It provides agents, handoffs between agents, guardrails and built-in tracing, and is available for Python and TypeScript.
Pricing: Open Source · Website (opens in new tab)
6. Pydantic AI
Pydantic AI is a Python agent framework from the team behind Pydantic. It is model-agnostic and type-safe end to end, with structured outputs, tools and dependency injection, and it integrates with Pydantic Logfire for observability. It is open source.
Pricing: Open Source · Website (opens in new tab)
7. Mastra
Mastra is an open-source TypeScript framework for building AI agents and apps. It provides memory, tools, MCP support, workflows and observability to take agents from prototype to production.
Pricing: Open Source · Website (opens in new tab)
8. Letta
Letta, from the team behind MemGPT, is a platform for stateful agents with long-term memory. Its agents remember across sessions and learn continuously over time. The core is open source, with a hosted cloud offering.
Pricing: Open Source · Website (opens in new tab)
Missing a tool? Submit it. Browse all frameworks.