Explanation
Reasoning planes
The specialized agents behind a review — each routed to the model that fits its job.
A codingassist.bot review isn't one prompt — it's a small team of specialized agents, each with a narrow job and each routed to the model that fits it. Context agents gather the evidence; review agents reason over it.
The roster
| Agent | Job | Model | Status |
|---|---|---|---|
| Scholar | RAG — top-K semantic retrieval across the whole repo | Embeddings | |
| Archaeologist | Reads git history & blame for intent | — | |
| Provost | Scores context, gates budget | Groq | |
| Critique | The core review — bugs, security, tech debt | Claude Sonnet | |
| Fixer | Proposes ready-to-apply patches (@codingassist fix) | Claude Sonnet | |
| Diplomate | Explains findings in the PR (@codingassist explain) | Groq | |
| Mentor | Learns from tenant feedback over time | — | planned |
How they fit the pipeline
Rendering diagram…
See the 3-stage pipeline for how context flows into the agents, and Acceptance-Criteria verification for the check CodeRabbit can't do.
Related
Was this page helpful?