Ruyi OpsBuddy: a seven-stage diagnosis loop that makes AI actually troubleshoot
Many ops AIs stop at chatting: they answer questions but cannot stand behind a conclusion. Ruyi OpsBuddy sets out to make AI troubleshoot like a senior ops engineer, leaving evidence at every step.
Seven stages
- Detect: take a one-sentence task or an alert and identify the assets and symptoms involved.
- Hypothesize: list possible root causes using expert standards and the skill library.
- Gather evidence: collect read-only evidence through connectors or the edge executor, and confirm or rule out hypotheses with data.
- Conclude: give a root-cause conclusion with a confidence level and recommendations.
- Act (with approval): write operations need approval by risk level (L0–L3), with multi-level approval chains and separation of duties.
- Verify: results are verified automatically; nothing is declared fixed without verification.
- Learn: feedback scores and experience distillation turn frequent troubleshooting paths into draft skills.
Experts and skills
OpsBuddy ships 74 skills written to one expert standard (decision trees, root-cause pattern tables, key metrics and thresholds, common mistakes and real cases) with a CI quality gate of at least 90 points; each of the 14 domain experts has a "first five minutes" checklist, operating red lines and delivery standards, and the incident war room lets several experts work together.
Useful without an LLM
With no model configured, the offline engine takes over: skill search, alert grouping and root cause, inspections, tickets, approvals and verification keep working; with models configured, Claude, Hunyuan, DeepSeek, Qwen, GLM, Ollama and vLLM are routed by task complexity, cost and latency. With "private models only" turned on, only local models are used.
Learn more on the Ruyi OpsBuddy product page or book a demo.