Free AI Agent Frameworks

An open-source framework for building apps on language models, with ready parts for prompts, tools, document search and agents. MIT licensed.

A Python framework for teams of AI agents, each with a role and a goal, working through tasks together. MIT licensed, with a paid hosted platform on top.

A TypeScript toolkit for adding AI chat, streaming and tool calls to web apps, with one API across many model providers. Apache 2.0 licensed.

Monitoring for AI agents. Records each run so you can replay it, see what it cost and find where it went wrong. The SDK is MIT licensed.

One of the first autonomous agent projects, now a platform for building agents that run tasks on their own. The platform code is under the Polyform Shield License and the rest is MIT, so it is not fully open source.

An open-source framework for building multi-agent systems and studying how agents work together. Apache 2.0 licensed.

Connects AI agents to hundreds of apps like Gmail, Slack and GitHub, and handles the logins and API calls. The SDK is MIT; the connections run through Composio's service.

A Stanford framework that swaps hand-written prompts for code. You describe what each step should do and it tunes the prompts for you. MIT licensed.

Secure cloud sandboxes where AI agents can run code without touching your machine. The SDK is open source; the sandboxes run on E2B's hosted service.

A memory layer for AI agents, so an assistant remembers users and past conversations between sessions. Apache 2.0, with a hosted version.

A multi-agent framework that gives separate agents roles like product manager, architect and engineer so they build software together. MIT licensed.

Ask questions about your own documents with an AI that can run fully offline, so no data leaves your machine. Apache 2.0 licensed.