2 KiB
2 KiB
Rust Engine API and Worker
Overview
- HTTP API (warp) under /api for file management and query lifecycle
- MySQL for metadata, Qdrant for vector similarity
- Background worker resumes queued work and re-queues stale InProgress jobs at startup
Environment variables
- DATABASE_URL: mysql://USER:PASS@HOST:3306/DB
- QDRANT_URL: default http://qdrant:6333
- GEMINI_API_KEY: used for Gemini content generation (optional in demo)
Endpoints (JSON)
-
POST /api/files (multipart)
- Form: file=@path
- Response: {"success": true}
-
GET /api/files/list
- Response: {"files": [{"id","filename","path","description"}]}
-
GET /api/files/delete?id=<file_id>
- Response: {"deleted": true|false}
-
POST /api/query/create
- Body: {"q": "text", "top_k": 5}
- Response: {"id": "uuid"}
-
GET /api/query/status?id=<query_id>
- Response: {"status": "Queued"|"InProgress"|"Completed"|"Cancelled"|"Failed"|"not_found"}
-
GET /api/query/result?id=<query_id>
- Response (Completed): { "result": { "summary": "Found N related files", "related_files": [ {"id","filename","path","description","score"} ], "relationships": "...", "final_answer": "..." } }
-
GET /api/query/cancel?id=<query_id>
- Response: {"cancelled": true}
Worker behavior
- Ensures Qdrant collection exists (dim 64, cosine)
- Re-queues InProgress older than 10 minutes
- Processing stages:
- Set InProgress
- Embed query text (demo now; pluggable Gemini later)
- Search Qdrant top_k (default 5)
- Join file metadata (MySQL)
- Gemini step: relationship analysis (strictly from provided files)
- Gemini step: final answer (no speculation; say unknown if insufficient)
- Persist result (JSON) and set Completed
- Checks for cancellation between stages
Local quickstart
- docker compose up -d mysql qdrant
- set env DATABASE_URL and QDRANT_URL
- cargo run
Notes
- Replace demo embeddings with real Gemini calls for production
- Add auth to endpoints if needed (API key/JWT)