GitHub

Patterns for teams shipping AI agents.

A curated reference for orchestration, safety, memory, evals, and human control. Built to help you choose the operating model before the implementation becomes expensive.

Contribute Pattern ↗ Find the Right Starting Point
154 Patterns
8 Categories
2 Domains
40 Field-tested
Curated Core

The patterns that inform architecture decisions early.

This library should feel opinionated. These entries are the fastest way to understand how agent systems break, how they recover, and where control needs to stay explicit.

Editorial Thesis

Reference material should help you make decisions, not just collect examples.

  • Bias toward patterns with operational consequences and real trade-offs.
  • Separate architecture, reliability, and safety instead of flattening them into tags.
  • Make it easy to move from exploration to a concrete build path.
Build Paths

Pick a route through the catalog.

The useful split is not “new vs updated.” It is whether you are choosing architecture, improving feedback loops, or hardening a risky system.

Build

Choose the architecture before you choose the model

Start with orchestration, context, and interface patterns that define how your agent actually behaves.

Orchestration & control Improve

Instrument the loop and harden the feedback path

Add evals, observability, and review loops early so the system can improve without guesswork.

Feedback loops & reliability Protect

Design guardrails as first-class product behavior

Treat safety, permissions, and approval flows as core UX, not compliance afterthoughts.

Security & safety
Dispatch

Follow the library as it matures.

Product teams do not need more vague AI commentary. They need a sharper feed of patterns, changes, and build guidance.