12 repositories for building AI agents that survive contact with a real product
The Agent Library is a free, curated set of 12 public GitHub repositories for building production AI agents, grouped into orchestration frameworks, TypeScript-native tooling, tool and browser integration, and the observability layer you need before launch. Every repository is open source and free to clone. Frenchy Digital adds the part the READMEs leave out: which one to pick, and what breaks when you do.
Free. 12 repositories, no paywall. Tip only if it helps.
A founder showed me a demo last month that worked. Actually worked. Built over about a weekend, maybe 300 lines, and it did the thing. Then he asked the only question that matters: what do we ship?
Somewhere between that demo and the App Store, the agent needs memory it can trust, a tool call that fails politely, a trace you can read at two in the morning, and a bill that doesn't quietly triple. None of that is in the tutorial he followed.
Here is the part most people get backwards. They think the bottleneck is picking the right framework, so they spend three weeks comparing them. It almost never is. The bottleneck is the boring layer underneath: what you log, what you test, what happens on the fourth retry. Teams that pick a framework in an afternoon and spend the saved fortnight on tracing ship sooner, roughly every time.
So one rule before you start clicking: choose a single orchestration framework and stay there for a month. Framework-hopping is the most expensive habit in this field right now. Every switch costs you the thing you were actually building, and they are converging anyway. The differences that look enormous in a comparison table mostly vanish once you have real users and real failure modes.
This is the list we hand clients. Twelve repositories, all public, none of them ours, which is the point. We've shipped on most of these and walked away from a couple, and the notes say which and why. That judgement is the only thing here you can't get off GitHub's trending page.
It's free. There's a tip button further down because a few people asked for one, and $15 is what we suggest, but nothing is behind it. Same email either way.
Orchestration: where the agent actually lives
Pick one. Picking two is how projects die. These are the four we have shipped against, and the honest reason you would choose each.
- langchain-ai/langgraph41,647· Python · MIT
Graphs, not chains. Reach for it when your agent has to branch, retry, and hand off to a human without losing its place.
- openai/openai-agents-python29,433· Python · MIT
The least ceremony between an idea and a running loop, as long as you are happy living inside one provider's tools.
- crewAIInc/crewAI58,551· Python · MIT
Role-playing agents with almost no boilerplate. Great for a demo on Friday; watch the token bill when it goes to production.
- microsoft/autogen60,984· Python · CC-BY-4.0
Conversation-first multi-agent work out of Microsoft Research. Strong ideas, a codebase that has been rewritten more than once.
- huggingface/smolagents29,328· Python · Apache-2.0
Agents that think in code, in about a thousand lines you can actually read. The best framework to learn from before you pick a big one.
- google/adk-python21,535· Python · Apache-2.0
Google's code-first agent toolkit. Worth a look if your data already lives in Google Cloud.
If your product is TypeScript, start here
Most agent tutorials assume Python. Most mobile and web products do not. These two close that gap.
- mastra-ai/mastra28,039· TypeScript · Elastic-2.0
Agents, workflows, memory and evals in TypeScript, designed to sit next to a Next.js app rather than behind a Python service.
- pydantic/pydantic-ai19,936· Python · MIT
Python, but typed end to end. The closest thing to the TypeScript developer experience if you are stuck on the Python side.
Tools: the part that makes an agent useful
A model with no hands is a chatbot. This is how you give it hands, and how you keep the connection standard instead of bespoke.
- modelcontextprotocol/servers90,335· TypeScript · MIT
Reference MCP servers. Read two of them and you will understand the protocol better than any explainer, this one included.
- browser-use/browser-use114,628· Python · MIT
When the system you need has no API and never will. Brittle by nature, so budget for it breaking.
Production: the boring half nobody posts about
The demo is the easy 20%. These two are what stand between a working prototype and something you would let a client near.
- langfuse/langfuse34,605· TypeScript · MIT / EE
Traces, evals and cost tracking you can self-host. The first thing we add to a project, before the second agent.
- anthropics/claude-cookbooks52,690· Jupyter Notebook · MIT
Runnable notebooks straight from Anthropic: tool use, retrieval, caching. Less theory, more copy-paste that works.
Star counts are a snapshot taken 2026-09-14, not live figures. Licences are as published by each project, so check them yourself before you ship commercially.
Want the whole list in your inbox?
Every repository above, with the notes, in one email you can search later, plus a note from me whenever we publish something new on AI and where mobile development is going. That's the only trade.
Tip if it helped. Only if it helped.
The library is free and stays free. You already have it, whether or not you click anything here. If it saved you an afternoon, you can name your own amount. It goes toward the time we spend testing these so you don't have to.
You can change the amount on the next screen. Secure checkout by Stripe, and we never see your card.
Questions people actually ask
Ready to Build Your Application?
Let's discuss your project. Our team will help you build solutions that transform your business.