AI Diff Review Service

Async HTTP service that reviews unified diffs and returns structured findings. Built with Python / FastAPI / asyncio. All state is in-memory.

Run locally

python -m venv .venv
.\.venv\Scripts\Activate.ps1      # Windows
pip install -r requirements.txt
$env:API_BEARER_TOKEN = "your-token"
$env:LLM_API_KEY = "your-gemini-key"     # optional; llm jobs fail gracefully without it
$env:LLM_MODEL = "gemini-3.5-flash-lite"
uvicorn app.main:app --port 8000

Tests: pytest -v (68 tests)

API

• GET /health, GET /spec — public
• POST /v1/reviews — submit a diff (bearer auth). Options: provider (mock|llm), maxFindings. Supports Idempotency-Key header.
• GET /v1/reviews/{jobId} — poll status/findings
• GET /v1/reviews/{jobId}/stream — SSE (status/finding/done), identical replay for finished jobs

Configuration (environment variables)

Var	Purpose	Default
API_BEARER_TOKEN	auth token for /v1/*	dev-token-change-me
LLM_API_KEY	Google Gemini API key	(unset → llm jobs fail gracefully)
LLM_MODEL	Gemini model id	gemini-3.5-flash-lite

Known limitations

• In-memory storage: jobs, cache, idempotency records, and rate-limit state are lost on restart. Acceptable for the scoring window; production would use Redis/Postgres.
• Single-process design: limits and caches are per-process.