JB SafeGrowth Agent

JB SafeGrowth Agent is a monorepo MVP for a financial advertising compliance workflow:

1. Content input
2. Redline risk review
3. Evidence lookup
4. Rewrite comparison
5. Approval package

The frontend and backend are deployed separately, but developed together in this repository.

Repository Layout

apps/frontend      Vercel-targeted React/Vite app
apps/backend       Render-targeted FastAPI app
infra/supabase     Supabase schema, migrations, and seed data
docs               Demo, deployment, handover, and diagram notes
.github/workflows  Separate frontend/backend CI skeletons

Local Backend

cd apps/backend
virtualenv --always-copy .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Optional local secrets live in apps/backend/.env. Leave it absent, or leave placeholder values such as replace-me, to run the deterministic fallback demo without Gemini or Supabase.

Health check:

curl http://localhost:8000/v1/health

Local Frontend

cd apps/frontend
npm ci
npm run dev

The Vite dev server defaults to http://localhost:5173.

Docker Compose

The backend service reads apps/backend/.env when it exists. If the file is absent, the app starts in deterministic fallback mode and must still support the demo workflow without Gemini or Supabase.

docker compose up --build backend
curl http://localhost:8000/v1/health

Do not use .env.example as a runtime env file. It is only a template for local setup.

Frontend Docker verification should also use lockfile-based installs:

cd apps/frontend
npm ci
npm run lint
npm run typecheck
npm run build

Secret Boundaries

• Frontend only receives backend API base URL values such as VITE_API_BASE_URL.
• GEMINI_API_KEY, OPENAI_API_KEY, SUPABASE_SERVICE_ROLE_KEY, and DATABASE_URL are backend-only.
• Set LLM_PROVIDER=openai_compatible with OPENAI_BASE_URL=http://host.docker.internal:18080/v1 when the backend container should use the shared local LLM server.
• Rule-based compliance detection must work without Gemini or Supabase.

Determinism

For stable analyze/rewrite responses set LLM_TEMPERATURE=0.0 (or 0.1) on the backend. Identical inputs are also served from an in-memory cache (TTL 15min). Bypass with ?refresh=1. See docs/deployment/README.md for the full env-var list.