docker-compose.yml

# Local containerized testing only (see docker/*.Dockerfile headers) - not
# what a later Render/Vercel deployment milestone will necessarily use.
# Two services, no orchestration beyond that: no Celery/Redis/Postgres/
# nginx, matching backend/pipeline_runner.py's own "no external limiter"
# design.

services:
  backend:
    build:
      context: .
      dockerfile: docker/backend.Dockerfile
    ports:
      - "8000:8000"
    environment:
      # Optional - falls back to Semantic Scholar's unauthenticated rate
      # limit if unset, same as running the backend locally without a
      # .env file (see core/config.py).
      - SEMANTIC_SCHOLAR_API_KEY=${SEMANTIC_SCHOLAR_API_KEY:-}
    volumes:
      # data/jobs.db + per-topic embedding/cluster caches survive a
      # `docker-compose down` / rebuild, same as the gitignored data/ dir
      # does for a local (non-Docker) run.
      - aporia-data:/app/data
      # SPECTER2 (core/embedding/specter_encoder.py) is downloaded from the
      # Hugging Face Hub the first time a job reaches the embedding stage,
      # not at image build time - persisting the cache means that download
      # only happens once across container restarts, not on every restart.
      - hf-cache:/root/.cache/huggingface

  frontend:
    build:
      context: frontend
      dockerfile: ../docker/frontend.Dockerfile
    ports:
      - "3000:3000"
    environment:
      # Overrides vite.config.ts's preview.proxy target (127.0.0.1, the
      # right value for local dev) to this compose network's backend
      # service hostname.
      - VITE_API_PROXY_TARGET=http://backend:8000
    depends_on:
      - backend

volumes:
  aporia-data:
  hf-cache: