Best GitHub Repositories for AI Agents in 2026: Platforms, Frameworks and Browser Agents

The GitHub repositories behind AI agents in 2026 — Dify, n8n, LangGraph, CrewAI, OpenHands and browser-use — grouped by what you are building, with live stars and maintenance status.

Short version: decide first whether you want a platform or a framework. Platforms (Dify, n8n) are faster to ship with; frameworks (LangGraph, CrewAI) give you control. Add browser-use if your agent must operate websites, and OpenHands if the agent writes code. Stars and push dates below are from the GitHub API on 18 September 2026.

GitHub repositories behind AI agents, by stars

Platforms: build agents visually

**n8n** (about 205,000 stars): workflow automation with native AI steps. Fair-code licence.
**Dify** (about 156,000): agentic workflows and RAG pipelines with a visual builder and model management.
**LobeHub** (about 83,000): a workspace for organising and running agents.

Frameworks: write agents in code

**LangGraph** (about 42,000, MIT): explicit, stateful agent graphs for production.
**CrewAI** (about 59,000, MIT): role-based multi-agent setups, quick to start.
**AutoGen** (about 61,000): Microsoft's multi-agent framework. Its last code push was in April 2026, so check the README for current guidance.

Agents that use the browser

**browser-use** (about 115,000, MIT): lets a model operate a real browser, for sites with no API.
**UI-TARS-desktop** (about 39,000): ByteDance's multimodal agent stack for operating desktop interfaces.

Agents that write code

**OpenHands** (about 88,000, MIT): autonomous coding in a sandbox.

Platform or framework: choosing an AI agent repository

How to choose

Who is building it?

- Ops or marketing team: a platform such as Dify or n8n.

- Engineers: a framework such as LangGraph or CrewAI.

What does the agent touch?

- APIs: any of them, through tools or MCP.

- Websites without APIs: browser-use.

- A codebase: OpenHands or a coding agent.

Check maintenance and licence.

- Look at the last push date and the licence file of the version you deploy.

For a deeper comparison of the frameworks alone, see our open-source agent frameworks guide.

Frequently asked questions

What is the difference between an agent platform and an agent framework?+

A platform, such as Dify or n8n, gives you a visual builder, hosting, logs and integrations, so you assemble agents without writing much code. A framework, such as LangGraph or CrewAI, is a library you import to write agents in Python or TypeScript, with full control and nothing hosted for you. Teams without dedicated engineers usually start with a platform.

Which AI agent framework should I learn in 2026?+

LangGraph is the common choice for production agents that need explicit control over steps, state and retries. CrewAI is quicker for multi-agent role setups. Both are MIT-licensed and actively developed. Learn one well rather than sampling five: the underlying ideas, tool calling, state and evaluation, transfer between frameworks far better than any specific API does.

Is AutoGen still maintained?+

Microsoft's AutoGen repository is still public with about 61,000 stars, but its most recent code push was in April 2026, much slower than the other frameworks here. Read the repository's README for the maintainers' current guidance on where new projects should go before building anything new on it.

Can AI agents use websites without an API?+

Yes. browser-use gives a language model a real browser: it reads the page, decides what to click or type, and drives the browser underneath. It is useful for internal dashboards and portals with no API. Each step is a model call, so long tasks get expensive, and sites that change their layout can break runs that a scripted scraper would handle.

Are these agent repositories free to use commercially?+

Most frameworks here are MIT or Apache-2.0 and free for commercial use. Platforms are more varied: n8n uses a fair-code licence that restricts offering it as a service, and several platforms keep some features in a paid enterprise edition. Check the licence file of the exact repository and version you deploy, not a summary written elsewhere.