Practical tools for AI agents that need to actually work.

Keats AI is a small operator-led lab built by Corbin Breton with an AI partner. We build open-source tools, skills, and operating patterns for OpenClaw and real-world agent systems.

This is not theory-first AI. Everything here comes from running agents, finding failure modes, and turning the useful parts into tools.

Multi-model PRD execution for OpenClaw.

Turn a staged PRD into delegated implementation with model routing, automatic escalation, scope enforcement, and validation gates.

clawhub install openforge

A compact operational playbook for building agent systems that persist, improve, and stay coherent over time. Covers the parts most people skip: identity, memory, trust, automation, and failure handling.

• Evidence before claims. Build from real use. Keep the receipts.
• Agents are operational systems. Identity, memory, trust, and constraint matter.
• Open source the sharp tools. Keep the useful parts public where possible.
• Small is fine. We'd rather ship working infrastructure than perform scale.

More skills are in development around identity, memory, ops hygiene, and autonomous improvement.

We are not interested in filling this page with vapor. When something is here, it's because it's real.

• Website: keats-ai.dev
• Blueprint: keats-ai.dev/blueprint
• OpenForge on ClawHub: clawhub.ai/bloodandeath/openforge
• Platform: OpenClaw

If you're using these tools and hit something sharp, open an issue. Good bug reports and real operator feedback are worth more than generic hype.