Summary

This repository curates practical, production-ready patterns for building autonomous and semi-autonomous AI agents. It addresses the gap between toy demos and real-world product complexities by cataloging repeatable tricks, workflows, and mini-architectures. The target audience includes engineers and researchers shipping AI agents, offering them proven methods to enhance agent capabilities in sensing, reasoning, and acting, thereby enabling faster and smarter agent development.

How It Works

The project defines an "agentic pattern" as a repeatable, agent-centric, and traceable technique backed by public references. Patterns are organized into distinct categories like Orchestration & Control, Context & Memory, Feedback Loops, Tool Use & Environment, UX & Collaboration, and Reliability & Eval. This structured approach allows users to quickly find and apply established solutions to common challenges in agent development, leveraging community-vetted methods rather than reinventing solutions.

Quick Start & Requirements

This repository is a curated list and does not require installation or direct execution. Users can browse the patterns directly from the README or contribute new patterns by forking the repository, adding a pattern file under the patterns/ directory, and opening a Pull Request. Links to inspiration sources are provided within the README.

Highlighted Details

• Categorized Patterns: Comprehensive organization across key agent development areas including Orchestration, Memory, Feedback Loops, Tool Use, UX, and Reliability.
• Production Focus: Emphasis on "real-world tricks, workflows, and mini-architectures" that bridge the gap between toy demos and production agents.
• Community-Driven: Encourages contributions via Pull Requests, fostering a collaborative environment for sharing and refining agentic techniques.
• Traceability: Each pattern is required to be backed by a public reference (blog, repo, paper), ensuring verifiability.

Maintenance & Community

The project is maintained through community contributions via GitHub Pull Requests. Inspiration is drawn from sources like "What Sourcegraph learned building AI coding agents" and the "Raising an Agent" video diary. Further details for contributors are available in CONTRIBUTING.md.

Licensing & Compatibility

The project is licensed under the Apache-2.0 license. This permissive license generally allows for commercial use, modification, and distribution, including use in closed-source projects, with standard attribution and notice requirements.

Limitations & Caveats

As a curated list, the repository's completeness is dependent on community contributions and may not cover every emerging agentic pattern. The definition of a "pattern" requires a public reference, potentially excluding valuable internal or undocumented techniques. The categorization is noted as fluid, suggesting potential overlap or evolving structures.