The Best Open-Source AI Agent Platforms in 2026
Self-hostable agent platforms and frameworks, compared honestly: which licenses are really open source, which projects are thriving, and which were archived this year.
Last updated Oct 8, 2026
Short answer: for a self-hosted visual platform, start with Dify (the most complete; Apache 2.0 with added conditions) or Langflow (MIT, the most permissive). Choose AutoGPT if you want to run a team of AI Experts rather than build individual flows; it's free to self-host and source-available under PolyForm Shield. Sim (Apache-2.0) is the actively developed newcomer. Developers who want a framework rather than a platform should look at LangGraph, CrewAI, Pydantic AI, smolagents or Microsoft Agent Framework. Skip Flowise: it was archived in August 2026.
Disclosure: we make AutoGPT, one of the platforms below. License terms were read from each project's license file, and GitHub star counts were pulled from the GitHub API on October 2, 2026.
Open-source AI agent platforms at a glance
| Project | What it is | License | GitHub stars |
|---|---|---|---|
| Dify | Visual platform for agents and agentic workflows | Apache 2.0 with added conditions | 158k |
| Langflow | Visual builder for AI agents and MCP servers | MIT | 155k |
| AutoGPT | AI Experts coordinated by Otto, plus a visual workflow builder | PolyForm Shield (platform), MIT (everything else) | 188k |
| Sim | Workspace to build, share and govern AI agents | Apache-2.0 | 30k |
| n8n | Workflow automation with AI agent nodes | Sustainable Use License (fair-code) | 206k |
| Activepieces | Workflow automation with AI agents | MIT core + commercial enterprise edition | 25k |
| LangGraph | Agent orchestration framework (Python, JavaScript) | MIT | 43k |
| CrewAI | Multi-agent framework (Python) | MIT | 59k |
| Microsoft Agent Framework | Agent framework; successor to AutoGen | MIT | 14k |
| Pydantic AI | Type-safe agent framework (Python) | MIT | 20k |
| smolagents | Minimal agent library (Python) | Apache-2.0 | 30k |
| Mastra | Agent framework (TypeScript) | Apache 2.0 core + source-available enterprise edition | 28k |
| OpenHands | Autonomous coding agent | MIT | 90k |
| Flowise | Visual LLM app builder (archived August 2026) | Apache 2.0 core + commercial enterprise code | 55k |
Stars rounded to the nearest thousand, from GitHub on October 2, 2026. Stars measure attention, not quality, but they're a useful signal of community size.
How we chose
Every platform here can run on your own infrastructure with its source code available. Check each linked repository's recent commits before choosing; activity changes over time. "Open source" gets used loosely in this category, so the license column matters. MIT and Apache-2.0 are permissive open-source licenses. Dify adds conditions to Apache 2.0. n8n's Sustainable Use License and AutoGPT's PolyForm Shield are source-available licenses that restrict certain commercial uses.
Open-source AI agent platforms
Dify
Dify bills itself as "the platform for production-ready agentic workflows." It combines a visual workflow and agent builder with RAG knowledge bases and model management, and you can deploy it "on cloud, VPC, or self-hosted." It's the default answer to "self-hosted agent platform" for good reason.
Watch out for: the license is Apache 2.0 plus conditions. You can't run Dify as a multi-tenant service for others without written permission, and you can't remove its logo or copyright from the frontend.
Langflow
Langflow is "a powerful tool to build and deploy AI agents and MCP servers," with a visual editor over Python components that you can customize. Because it's MIT-licensed, you can embed it in your own product, and you can "deploy yourself or sign up for a free cloud account."
Watch out for: it's a builder more than an operations platform, so plan how you'll host, monitor and govern what you build.
AutoGPT
AutoGPT launched in 2023 as the open-source autonomous agent project that helped start this category, and it's still one of the most-starred AI repositories on GitHub. Today's AutoGPT platform gives you a team of AI Experts (specialists with a role, instructions, skills and workflows) coordinated by Otto, your Head of AI. You start from a conversation rather than a canvas, connect the tools an Expert needs, review its work and turn recurring jobs into routines. The visual workflow builder is there when you want to define each step, and Experts can build and run workflows for you.
Self-hosting is free; you supply the infrastructure and model access. One honest caveat: the platform code is source-available under PolyForm Shield, which allows any use except providing a product that competes with AutoGPT, so it isn't OSI open source. The classic AutoGPT agent and everything outside the platform folder are MIT.
Sim
Sim describes itself as "the open-source AI workspace where companies build, distribute, and govern AI agents." It's Apache-2.0, and you can "run Sim in your own cloud with Docker or Kubernetes." It's younger than Dify or Langflow and under active development.
n8n
n8n is the most-starred project on this list and has deep AI agent tooling, but it's a workflow automation platform first: agents run as nodes inside workflows you design. Its license is fair-code rather than open source. It's free to self-host for internal business use, with limits on reselling it or offering it as a service. If those limits are a problem, see our n8n alternatives.
Activepieces
Activepieces pairs an MIT-licensed workflow builder with "AI agents with enterprise control," and runs "with Docker or Kubernetes on your own infrastructure." Its 750+ integrations make it a practical pick when your agents need to touch business apps.
Open-source agent frameworks for developers
If you're building agents into your own product, a framework gives you more control than a platform. The strongest options in 2026:
- LangGraph (MIT, 43k stars): stateful, long-running agent graphs in Python or JavaScript, and a common default for production agent orchestration.
- CrewAI (MIT, 59k): multi-agent "crews" with roles, goals and tasks, in Python.
- Microsoft Agent Framework (MIT, 14k): Microsoft's successor to AutoGen, which is now in maintenance mode.
- Pydantic AI (MIT, 20k): type-safe Python agents from the team behind Pydantic.
- smolagents (Apache-2.0, 30k): Hugging Face's minimal library for agents that write and run code.
- Mastra (Apache 2.0 core, 28k): "the leading TypeScript agent framework," in its own words.
- Haystack (Apache-2.0, 27k): deepset's framework for retrieval-heavy pipelines and agents.
- OpenAI Agents SDK (MIT, 30k) and Claude Agent SDK (MIT, 8k): vendor SDKs that are open source but built first for their makers' models.
Open-source specialist agents
- OpenHands (MIT, 90k): an autonomous coding agent you can self-host. The repository now lives under the OpenHands organization.
- Letta (Apache-2.0, 25k) and Mem0 (Apache-2.0, 66k): memory layers for agents that need to remember across sessions.
Archived or winding down in 2026
- Flowise was archived on GitHub on August 13, 2026, after a code freeze in July, and reached end of life on August 31. There will be no further fixes or security updates. Langflow and Dify are the natural replacements.
- AutoGen is in maintenance mode: it "will not receive new features or enhancements and is community managed going forward." Microsoft points new users to Microsoft Agent Framework.
- OpenAI Agent Builder isn't open source, but if you're looking for a self-hosted replacement, note that it's scheduled to shut down on November 30, 2026.
Open source vs source-available: what the licenses allow
- MIT and Apache-2.0 (Langflow, Sim, LangGraph, CrewAI and most frameworks): use, modify, embed and resell freely, with attribution.
- Apache 2.0 with conditions (Dify): open-source terms plus two restrictions. You can't run it as a multi-tenant service without permission, and the frontend logo stays.
- Open core (Activepieces, Mastra, Flowise): a permissive core, with enterprise features under a commercial license.
- Sustainable Use License (n8n): free for internal business, non-commercial or personal use; no reselling it or hosting it for others.
- PolyForm Shield (the AutoGPT platform): any use except providing a product that competes with AutoGPT.
Frequently asked questions
What is the best open-source AI agent platform?
For a visual, self-hosted platform, Dify is the most complete and Langflow the most permissive (MIT). If you want AI Experts that do recurring work rather than flows you build, AutoGPT self-hosts for free (source-available). For developers, LangGraph and CrewAI are the most widely used frameworks.
Is AutoGPT still open source?
Partly. The classic AutoGPT agent and the code outside the platform folder are MIT-licensed. The AutoGPT platform, which runs the Experts and the workflow builder, is source-available under PolyForm Shield: free to self-host and modify for any purpose except providing a product that competes with AutoGPT. All of it is on GitHub.
Is Dify open source?
Dify uses Apache 2.0 with added conditions: you can't operate it as a multi-tenant service for others without written permission, and you can't remove its logo or copyright from the frontend. For most internal deployments, those conditions won't matter.
What happened to Flowise?
The Flowise team froze the code in July 2026, archived the GitHub repository on August 13, and set end of life for August 31, 2026. Langflow and Dify are the closest open-source replacements.
What's a self-hosted alternative to OpenAI's Agent Builder?
OpenAI's Agent Builder is scheduled to shut down on November 30, 2026, and OpenAI recommends its Agents SDK or ChatGPT Workspace Agents. If you'd rather run a visual builder on your own servers, Dify, Langflow, Sim and AutoGPT all self-host.
Sources
Checked Oct 2, 2026 against each vendor's own site or repository.
All product names, logos, and trademarks are the property of their respective owners. Comparisons are based on publicly available information as of the date shown and imply no affiliation with or endorsement by the companies named.
See what an Expert can take off your plate
Tell Otto what needs doing, or choose an AI Expert and give them the context for the job. See free trial terms.