Kyūmei (究明) is an open-source, local-first AI diagnostic agent designed to investigate hardware and device problems through an evidence-grounded reasoning workflow.

Instead of treating an LLM as a simple question-and-answer interface, Kyūmei gives it a structured diagnostic process through a reusable Agent Skill. The agent extracts evidence from a user’s symptoms, forms possible causes, links each hypothesis back to supporting evidence, proposes safest-first troubleshooting steps, and defines how each hypothesis should be verified.

The system then uses deterministic validation around the model rather than blindly trusting its output. Responses are validated against a JSON Schema, evidence-to-cause relationships are verified, unsupported hypotheses can be rejected, and safety guardrails handle situations involving batteries, smoke, burning smells, exposed power, and other potentially hazardous conditions.

Kyūmei is designed to run locally through Ollama using an open-weight model, keeping the diagnostic pipeline usable without relying on a closed AI API.

The project combines an open-weight AI model, a reusable Agent Skill, structured agent workflow, deterministic verification, and safety guardrails into a small open-source diagnostic system.

The goal is not simply to make an AI that can diagnose hardware, but to explore how an open-weight model can be made more reliable and useful through the architecture surrounding it.