12-factor-agents

Why we included this project

Most agent frameworks sell you a library, but this project treats LLM application engineering as a set of decisions you make yourself. It lays out twelve practices, each in its own chapter: own your prompts and context window, treat tools as structured outputs, unify execution and business state, and keep agents as stateless reducers. The value is that it gives you a vocabulary and a checklist to work through before you pick any framework, which is exactly what teams designing their own agent architecture need. There is also a scaffolding tool that generates a project already following the principles, so you can test the guidance instead of just reading it. If you are deciding how to structure a production agent, or you want one shared reference to get your team agreeing on what reliable means, this is a good place to start.