Guide · Updated October 2026
Open source
AI agents
The main open source and fair-code agent frameworks, what their licenses allow, when to self-host and the guardrails every business agent needs.
Guide · Updated October 2026
The main open source and fair-code agent frameworks, what their licenses allow, when to self-host and the guardrails every business agent needs.
We build agents on several of these tools and are not paid by any of them. Licenses and project status change: read each project's LICENSE file and repository before you commit. Checked on 4 November 2026.
An AI agent is software that uses a language model to decide which steps to take, calls tools such as your CRM, calendar or database, and checks the result before moving on. An open source agent framework gives you the code to build and run that yourself, on your own servers if you want, with the model of your choice.
| Project | Type | License (check the repo) | Best for |
|---|---|---|---|
| LangGraph | Python and JavaScript library from LangChain | MIT | Developers who want full control of agent state and steps |
| CrewAI | Python library for teams of agents with roles | MIT for the framework; a paid platform is separate | Multi-agent tasks like research, then write, then review |
| AutoGen and AG2 | Python multi-agent frameworks | Microsoft AutoGen code MIT; the AG2 community fork Apache 2.0 | Research and multi-agent conversations |
| n8n | Visual workflow tool with AI agent nodes | Fair-code Sustainable Use License, not OSI open source; free to self-host for internal use | Business workflows where agents call your existing apps |
| Dify | Visual platform for AI apps and agents | Modified Apache 2.0 with limits on multi-tenant SaaS and branding | Chat assistants and internal tools with a visual builder |
| Flowise | Visual builder for LLM flows | Apache 2.0 | Reported as archived in 2026 after Workday acquired the team in 2025; check the repository before starting anything new |
MIT and Apache 2.0 let you use, change and ship the code commercially. Fair-code and modified licenses, like n8n's and Dify's, are free for many uses but restrict offering the tool itself as a hosted service. If you plan to sell an agent product built on one of them, read the license or ask a lawyer.
This pattern works the same in n8n, LangGraph or CrewAI. The framework changes how you build it, not what it does.
For most small businesses, the fastest useful agent is a workflow agent in n8n that reads an incoming email or form, looks up the customer, drafts a reply and waits for approval. Developers who need full control usually pick LangGraph. We build both through our AI automation service. Not sure which fits? Ask us.
FAQs
Widely used options include LangGraph, CrewAI and AutoGen or its AG2 fork for code-first builds, and n8n or Dify for visual builders. Check each license, since n8n and Dify are fair-code or modified licenses rather than OSI open source.
Yes. Most of these frameworks can be self-hosted, and you can point them at a hosted model API or a model you run yourself. You are then responsible for updates, security and backups.
n8n is fair-code under its Sustainable Use License. You can self-host it for free for your own internal business use, but the license is not OSI-approved open source and limits offering n8n as a service to others.
Reports in 2026 say the Flowise repository was archived after Workday acquired the team in 2025. Check the GitHub repository before building anything new on it.
At high volume they often are, because you pay for servers instead of per run. At low volume, a hosted platform is often cheaper once you count the time to maintain a server.
Tell us what it should do and where your data lives. We will recommend a framework and send a written quote.
Ask us