12-Factor Agents is an open-source guide to designing reliable LLM-powered applications using practical software-engineering principles rather than treating an agent as an unconstrained prompt-and-tools loop. Inspired by the Twelve-Factor App methodology, it argues that strong production agents are mostly well-structured software with carefully chosen LLM steps.

The project presents twelve design patterns covering areas such as natural-language-to-tool-call conversion, prompt and context ownership, tool design, state management, control flow, error handling, human interaction, and small focused agents. The material is written as an engineering playbook for teams moving from demos toward systems that are predictable, debuggable, and maintainable.

12-Factor Agents is especially useful for developers and technical leaders evaluating agent architectures or building their own frameworks. It is less a ready-made agent platform than a set of reusable principles for deciding where models should act autonomously and where deterministic application code should remain in control.