OMK Documentation
Observe. Measure. Know. Make every knowledge change in your AI application evidence-backed.
Browse by audience or category. For Chinese docs see 简体中文 index.
I want to use omk
- Who omk is for — the user, problem, and first ship/no-ship workflow
- Quickstart — first eval in 5 minutes
- Install the omk Agent Skill — agent-driven onboarding with
omk install omk-agent-skill - Use OMK in a Node.js service — choose a scorer, connect your service, and read the results
- CLI reference
- Core API (advanced integration)
- Eval sample format
- Executors
- Artifact & variant layout
- Comparison with 7 tools
- Glossary
How-to guides
- Run doctor checks
- Evaluate an agent (project-level runtime context)
- Auto-improve a skill
- Observe production traces
- Compose the OMK MCP integration
- Reproduce Codex parent/subagent observation
- Use non-Claude models (GLM / Qwen / DeepSeek / Moonshot / Ollama)
I want to understand how it works
- How OMK understands knowledge
- Who omk is for — why doctor / eval are the pre-ship trunk and observe is post-ship feedback
- The three stages: doctor / eval / observe
- Architecture
- Statistical rigor
- Scoring pipeline