The pattern book for agents that actually ship

What it does The repository collects agentic AI patterns—mini-architectures, workflows, and operational tricks—that teams use to keep autonomous or semi-autonomous agents working in production. Each entry is backed by a public reference and judged by whether it improves how an agent senses, reasons, or acts. The list is auto-generated from markdown files in a patterns/ folder and surfaced through a companion website with filtering, comparison, and graph visualization.

The interesting bit Instead of celebrating one-off prompts, the project treats agent engineering as infrastructure: every pattern must be repeatable across teams and traceable to a real blog post, talk, or paper. That makes it closer to a field guide than a hype list.

Key highlights

• Patterns span eight categories, including Context & Memory, Feedback Loops, Orchestration & Control, and Reliability & Eval.
• Concrete entries include Agent Circuit Breaker, Budget-Aware Model Routing with Hard Cost Caps, and Memory Synthesis from Execution Logs.
• The companion website offers a Pattern Explorer, side-by-side Compare Tool, Decision Explorer, and graph visualization of pattern relationships.
• The entire catalogue is auto-generated from the patterns/ folder, so the README and site stay in sync.

Verdict Worth bookmarking if you are building or operating LLM agents and want to move beyond basic prompt-and-tool recipes. Skip it if you are looking for a framework or library—this is strictly a reference catalogue and website.