Libraries and platforms for building your own agents.
21 tools
Guides: Best frameworks for building agents, Best open-source AI tools to run yourself.
Agent frameworks
Agent-Native is an open source TypeScript framework for building agents that share actions with a user interface. You define a capability once, and the agent, the app, and other clients call the same code.
Agent frameworks
AutoGen is Microsoft's multi-agent framework. It is in maintenance mode, so new work is directed to Microsoft Agent Framework, but the library and docs remain published.
Agent frameworks
Browser Use lets an agent operate a real browser. The open-source library is Python, and the cloud SDK supports Python and TypeScript.
Agent frameworks
Composio connects an agent to apps such as GitHub, Slack, and Gmail, and handles sign-in for those tools. The SDK is open source, and the hosted service has a free tier.
Agent frameworks
CrewAI lets you define agents with roles and tasks that they work through together. The framework is open source, with a paid cloud for running crews.
Agent frameworks
Dify is an open source platform for building AI apps and agents with a visual editor. You can self-host it or use Dify Cloud.
Agent frameworks
Flowise is an open source visual builder for model workflows. You connect nodes for models, tools, and data, then call the result from an API.
Agent frameworks
The Agent Development Kit is Google's open source framework for building, evaluating, and deploying agents. It ships in several languages.
Agent frameworks
Haystack is an open source framework from deepset for search and question answering. You compose pieces such as retrievers and generators.
Agent frameworks
Hugging Face hosts models, datasets, and Spaces where people demo AI apps. Browsing the hub is free, and many of the libraries around it are open source.
Agent frameworks
LangChain is a library for building apps on language models. It covers prompts, retrieval, and agents in Python and JavaScript.
Agent frameworks
LangGraph is a library for agents that follow a graph of steps, with pauses and retries. It comes from the LangChain team.
Agent frameworks
Letta, formerly MemGPT, is an open source platform for agents that keep memory over time. You can run it yourself or use the hosted service.
Agent frameworks
LlamaIndex connects documents and data to language models. People use it to build search and question answering over private files.
Agent frameworks
Mastra is a TypeScript framework for agents and workflows. It is meant to live inside a Node or Next.js app.
Agent frameworks
Microsoft Agent Framework is Microsoft's current library for building agents in Python and .NET. New projects are pointed here instead of AutoGen or Semantic Kernel.
Agent frameworks
Ollama runs open models on your own computer. You pull a model and chat with it, or call it from other tools on localhost.
Agent frameworks
The OpenAI Agents SDK is a Python library for agents that call tools, hand work to other agents, and check results. It is open source.
Agent frameworks
PydanticAI is a Python agent framework from the Pydantic team. Type hints describe the tools and the shape of the model's output.
Agent frameworks
Semantic Kernel is Microsoft's library for connecting models to your own code. It is still published, and Microsoft now points new agent projects to Microsoft Agent Framework.
Agent frameworks
The AI SDK is Vercel's TypeScript toolkit for calling models, streaming text, and building agents. It works with React, Next.js, and other web frameworks.