Kav AI Server Handbook
The agent pipeline behind the Kav AI Assistant.
Introduction
The Kav AI Server (kavai_server) turns a natural-language question into a structured, evidence-backed answer through a Google ADK SequentialAgent pipeline (query_resolution). Every turn flows through three primary agents, with a vision AgentTool available for surface analysis:
The agents
| Stage | Agent | Responsibility |
|---|---|---|
| 1 | Resolver Classifier | Classify intent, extract structured filters, resolve conversational ambiguity and relative dates, plan downstream sub-tasks. |
| 2 | Database Specialist | Translate the resolved query into SQL against Supabase; coordinate surface analysis. |
| — | RGB Surface Analyzer | Vision AgentTool wired to the database specialist: per-image surface-condition detection on full-resolution RGB images. |
| 3 | Report Generator | Synthesize the retrieved records into a clear, structured Markdown report under a strict “no analysis” mandate. |
Source
These pages are the agent design documents maintained with the kavai_server backend (backend/kavai_server/docs/design/agents). They describe each agent’s role, contract, prompt inputs, tools, and state interactions.