Most frameworks on a typical shortlist are open source under permissive licenses. LangGraph is MIT licensed, and so are the OpenAI Agents SDK and Pydantic AI. CrewAI's LICENSE grants permission, free of charge, to any person who obtains a copy of the software. Agno ships Apache 2.0, as does Google ADK's repository. Mastra is Apache 2.0 outside ee/ directories, and n8n's Sustainable Use License limits use to internal business, non-commercial or personal purposes.
What is the difference between permissive and source-available licenses?
A permissive license lets you use, change, ship and sell the code, and mostly asks that you keep the copyright and license notice in your copies. MIT and Apache 2.0 are the two permissive licenses that come up on this shortlist.
A source-available license lets you read the code but attaches conditions to how you run or distribute it, such as limits on purpose, on who may build with it, or on use in production. Mastra's docs call code in any ee/ directory source-available under the Mastra Enterprise License, which needs a valid enterprise license for production use. The difference shows up when you sell a product, resell a service or host software for clients. Microsoft Agent Framework is outside the scope of this guide, so read its repository LICENSE before you commit to it.
Which multi-agent frameworks use the MIT license?
LangGraph, the OpenAI Agents SDK and Pydantic AI each ship an MIT license file. The text grants the rights to use, copy, modify and sell copies, and its one condition is that the copyright and permission notice stay in all copies or substantial portions.
CrewAI's LICENSE carries the same wording. It grants permission to deal in the Software without restriction, including the rights to use, copy, modify, merge, publish, distribute, sublicense and sell copies.
Contributing is a separate question from using. The OpenAI Agents SDK README says pull requests are limited to repository collaborators, so outside users report bugs and request features through GitHub issues.
Which frameworks use Apache 2.0, and what does it add?
Agno and Google ADK both use the Apache License 2.0. Agno's README says Agno is distributed under Apache 2.0. Google ADK's LICENSE file carries the same terms, ending with the appendix on applying the Apache License.
Apache 2.0 is also permissive, and it spells out more conditions in writing than a short MIT file does.
- Each contributor grants a royalty-free, irrevocable patent license covering its own contributions.
- If you start patent litigation claiming the Work infringes, your patent licenses under it terminate as of the date you file.
- Anyone you redistribute to must receive a copy of this License along with the code.
- Files you change must carry prominent notices stating the change when you distribute them.
How does Mastra's enterprise license work?
Mastra runs a dual-license model: Apache 2.0 for most of the codebase and the Mastra Enterprise License for enterprise features in ee/ directories. Those features require a valid enterprise license in production and can be used freely while you develop and test.
The trigger is configuration. Mastra says the need for a license depends on which features you configure, not on which packages you install. You set the key in the MASTRA_LICENSE_KEY environment variable, and Mastra also reads the older MASTRA_EE_LICENSE variable.
Without a valid key in production, a server configured with RBAC, FGA or Agent Builder does not start, and the error names the feature. A Studio sign-in screen that uses a third-party provider is hidden instead, while authentication on API routes keeps working.
What does n8n's Sustainable Use License allow?
n8n is the most restrictive license covered here. Most of the repository is under the Sustainable Use License. Files with ".ee." in the filename or ".ee" in the directory name are excluded from it, and using them requires a valid n8n Enterprise License.
The limitations clause allows use or modification only for internal business purposes or for non-commercial or personal use. You may give the software to others only free of charge and for non-commercial purposes.
The license FAQ draws the line for agent builders. Among the things you cannot do, it lists letting external end users build their own workflows through your product, including through an AI agent acting for them. Consulting work is allowed, and the FAQ says clients receive outputs, not access.
Which paid platforms sit next to the open-source libraries?
Several vendors keep the library free and charge for a hosted platform around it.
- LangChain's pricing page lists langgraph under Open Source as an agent framework for building agents with low-level control. LangSmith is the paid platform: Developer is $0 per seat with a maximum of one seat, and Plus is $39 per seat per month with unlimited seats.
- CrewAI's Basic plan is free with 50 workflow executions a month, and Enterprise is priced as Custom.
- Agno's pricing page says the runtime is free and open source. Pro costs $150 per month with three team seats and one live connection.
- Mastra lists a Free plan at $0 per month, Starter at $25 per month and Team at $250 per month. RBAC, SSO and audit logs are part of the Enterprise plan.
- The Pydantic AI README describes Pydantic Logfire as the AI engineering platform that sees your whole app, and it says Logfire has a free tier that needs no credit card. Pydantic AI's instrumentation is plain OpenTelemetry, so a backend you already run also works.
For how these frameworks differ beyond licensing, see LangGraph vs CrewAI and Agno vs CrewAI.
How do you pick a license for the way you will ship?
Start from how the code will leave your hands. For a product you sell or embed, the MIT and Apache 2.0 libraries give you the most room, as long as you keep the notices and, for Apache 2.0, mark the files you changed.
With Mastra, list the features your production config turns on before launch, because the enterprise requirement follows configuration. Our Mastra vs LangGraph comparison covers the frameworks themselves.
Plans to put n8n inside something customers pay for need a check against the FAQ examples first. For shipping n8n to customers as part of a paid product, the FAQ says to confirm your architecture with license@n8n.io before launch. The n8n vs LangGraph comparison covers product differences.