12 factor agents

Report Abuse

Basic Information

A public guide and reference of engineering principles for building reliable, production-ready LLM-powered applications. Inspired by 12 Factor Apps, the repository defines twelve practical factors covering topics such as converting natural language to tool calls, owning prompts and context windows, treating tools as structured outputs, unifying execution and business state, simple launch/pause/resume APIs, human contact via tool calls, owning control flow, compacting errors into context, small focused agents, triggering from anywhere, and making agents stateless reducers. It is aimed at engineers and technical founders who are integrating agentic behavior into existing products and want modular, maintainable patterns rather than wholesale framework rewrites. The repo collects factor writeups, diagrams, visual navigation, contributors notes, and related resources to help teams design and audit agent architectures.

Links

Categorization

App Details

Features
Detailed writeups for each of the twelve factors with individual pages describing rationale and implementation guidance. Visual navigation and diagrams illustrate agent loops, DAGs, control flow, and state patterns. The guide provides design patterns and practical advice for integrating modular agent concepts into existing products rather than forcing a full framework migration. It includes an appendix and honorable mentions, links to talks and videos, contributor-curated resources, diagrams and images for quick reference, versioning information with a v1.1 branch draft, and clear licensing: content under CC BY-SA 4.0 and code under Apache 2.0. Examples focus on tool calling, state management, error compaction, and small focused agents.
Use Cases
Helps engineers and founders design LLM applications that reach production quality by providing a concise, opinionated checklist of principles to evaluate and apply. The guide clarifies which responsibilities belong to LLMs versus deterministic code, how to structure prompts and manage context windows, how to unify execution and business state, and how to expose simple APIs for launch, pause, and resume workflows. It promotes modular adoption of agentic ideas into existing stacks, reducing rework and surface area for failure, and distills lessons learned from many teams and frameworks. Guidance is language agnostic with notes about TypeScript usage and applicability to other ecosystems, and the repo points to community resources for implementation.

Please fill the required fields*