docker-compose.yml

version: "3.9"

# ── Report Genius AI — Production Stack ──────────────────────────────────────
#
# Services:
#   api              — FastAPI + Uvicorn (FAISS by default; optional Qdrant for async pipeline)
#   qdrant           — Optional vector DB when VECTORSTORE_BACKEND=qdrant
#   temporal         — Optional durable workflow server (ENABLE_TEMPORAL_WORKFLOW=true)
#   temporal-worker  — Runs worker.py when Temporal is enabled
#
# Quick start:
#   cp .env.example .env          # set OPENAI_API_KEY and TENANT_SECRET_KEY
#   docker compose up --build -d
#   open http://localhost:8000
#
# To use PostgreSQL instead of SQLite, set DATABASE_URL in .env:
#   DATABASE_URL=postgresql+asyncpg://user:pass@postgres:5432/reportgenius
#
# Phase 2 + 3 (scale):
#   SCALE_OPTIMIZATION_PROFILE=true REDIS_URL=redis://redis:6379/0 \
#     docker compose --profile redis --profile jobs up -d
# ─────────────────────────────────────────────────────────────────────────────

services:

  redis:
    image: redis:7-alpine
    container_name: report_genius_redis
    restart: unless-stopped
    ports:
      - "6379:6379"
    profiles:
      - redis

  jobs-worker:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: report-genius-ai:latest
    container_name: report_genius_jobs_worker
    restart: unless-stopped
    command: ["python", "jobs_worker.py"]
    environment:
      OPENAI_API_KEY: "${OPENAI_API_KEY}"
      TENANT_SECRET_KEY: "${TENANT_SECRET_KEY:-change-me-in-production}"
      DATABASE_URL: "${DATABASE_URL:-sqlite+aiosqlite:////data/dev.db}"
      FAISS_INDEX_PATH: "/data/faiss_index"
      UPLOAD_DIR: "/mnt/data/uploads"
      CACHE_DIR: "/tmp/section_cache"
      REDIS_URL: "${REDIS_URL:-redis://redis:6379/0}"
      ENABLE_JOB_QUEUE: "true"
      SCALE_OPTIMIZATION_PROFILE: "${SCALE_OPTIMIZATION_PROFILE:-true}"
      ENABLE_ASYNC_PIPELINE: "${ENABLE_ASYNC_PIPELINE:-false}"
      ENABLE_SPECULATIVE_EXECUTOR: "${ENABLE_SPECULATIVE_EXECUTOR:-false}"
      ENABLE_PROMPT_CACHING: "${ENABLE_PROMPT_CACHING:-false}"
      VECTORSTORE_BACKEND: "${VECTORSTORE_BACKEND:-faiss}"
      QDRANT_URL: "${QDRANT_URL:-http://qdrant:6333}"
      GENERATION_STALE_SWEEP_SECONDS: "${GENERATION_STALE_SWEEP_SECONDS:-120}"
      GENERATION_TIMEOUT_SECONDS: "${GENERATION_TIMEOUT_SECONDS:-3600}"
    volumes:
      - uploads:/mnt/data/uploads
      - section_cache:/tmp/section_cache
      - db_data:/data
      - faiss_index:/data/faiss_index
    depends_on:
      redis:
        condition: service_started
    profiles:
      - jobs

  qdrant:
    image: qdrant/qdrant:v1.12.5
    container_name: report_genius_qdrant
    restart: unless-stopped
    ports:
      - "6333:6333"
    volumes:
      - qdrant_data:/qdrant/storage
    profiles:
      - qdrant

  api:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: report-genius-ai:latest
    container_name: report_genius_api
    restart: unless-stopped
    ports:
      - "8000:8000"
    environment:
      # Pin PORT=8000 so the container's CMD listens on 8000 and the existing
      # `localhost:8000` URL keeps working. The Dockerfile defaults to 7860
      # (Hugging Face Spaces convention) when this env var is absent.
      PORT: "8000"
      OPENAI_API_KEY: "${OPENAI_API_KEY}"
      TENANT_SECRET_KEY: "${TENANT_SECRET_KEY:-change-me-in-production}"
      DATABASE_URL: "${DATABASE_URL:-sqlite+aiosqlite:////data/dev.db}"
      FAISS_INDEX_PATH: "/data/faiss_index"
      UPLOAD_DIR: "/mnt/data/uploads"
      CACHE_DIR: "/tmp/section_cache"
      MAX_CONTEXT_TOKENS: "${MAX_CONTEXT_TOKENS:-400}"
      MAX_OUTPUT_TOKENS: "${MAX_OUTPUT_TOKENS:-300}"
      RETRIEVAL_TOP_K: "${RETRIEVAL_TOP_K:-10}"
      RERANK_TOP_N: "${RERANK_TOP_N:-3}"
      MAX_CONCURRENT_INGESTS: "${MAX_CONCURRENT_INGESTS:-4}"
      RATE_LIMIT_GENERATE_RPM: "${RATE_LIMIT_GENERATE_RPM:-20}"
      RATE_LIMIT_READ_RPM: "${RATE_LIMIT_READ_RPM:-120}"
      DEV_MODE: "false"
      SCALE_OPTIMIZATION_PROFILE: "${SCALE_OPTIMIZATION_PROFILE:-false}"
      ENABLE_ASYNC_PIPELINE: "${ENABLE_ASYNC_PIPELINE:-false}"
      ENABLE_SPECULATIVE_EXECUTOR: "${ENABLE_SPECULATIVE_EXECUTOR:-false}"
      ENABLE_PROMPT_CACHING: "${ENABLE_PROMPT_CACHING:-false}"
      ENABLE_TEMPORAL_WORKFLOW: "${ENABLE_TEMPORAL_WORKFLOW:-false}"
      ENABLE_HYBRID_RETRIEVAL: "${ENABLE_HYBRID_RETRIEVAL:-true}"
      SEMANTIC_CACHE_ENABLED: "${SEMANTIC_CACHE_ENABLED:-true}"
      AGENTIC_INSPECTOR_WHEN_NOTES_ONLY: "${AGENTIC_INSPECTOR_WHEN_NOTES_ONLY:-false}"
      NOTES_ONLY_GENERATION: "${NOTES_ONLY_GENERATION:-true}"
      GENERATION_TIMEOUT_SECONDS: "${GENERATION_TIMEOUT_SECONDS:-3600}"
      GENERATION_STALE_SWEEP_SECONDS: "${GENERATION_STALE_SWEEP_SECONDS:-120}"
      TEMPORAL_HOST: "${TEMPORAL_HOST:-temporal:7233}"
      TEMPORAL_NAMESPACE: "${TEMPORAL_NAMESPACE:-default}"
      VECTORSTORE_BACKEND: "${VECTORSTORE_BACKEND:-faiss}"
      QDRANT_URL: "${QDRANT_URL:-http://qdrant:6333}"
      MAX_CONCURRENT_LLM_CALLS: "${MAX_CONCURRENT_LLM_CALLS:-10}"
      REDIS_URL: "${REDIS_URL:-}"
      ENABLE_JOB_QUEUE: "${ENABLE_JOB_QUEUE:-false}"
    volumes:
      - uploads:/mnt/data/uploads
      - section_cache:/tmp/section_cache
      - db_data:/data
      - faiss_index:/data/faiss_index
    depends_on:
      qdrant:
        condition: service_started
        required: false
      temporal:
        condition: service_started
        required: false
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 20s

  temporal:
    image: temporalio/auto-setup:1.25.2
    container_name: report_genius_temporal
    restart: unless-stopped
    ports:
      - "7233:7233"
    environment:
      - DB=sqlite
    profiles:
      - temporal

  temporal-worker:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: report-genius-ai:latest
    container_name: report_genius_temporal_worker
    restart: unless-stopped
    command: ["python", "worker.py"]
    environment:
      OPENAI_API_KEY: "${OPENAI_API_KEY}"
      TENANT_SECRET_KEY: "${TENANT_SECRET_KEY:-change-me-in-production}"
      DATABASE_URL: "${DATABASE_URL:-sqlite+aiosqlite:////data/dev.db}"
      FAISS_INDEX_PATH: "/data/faiss_index"
      UPLOAD_DIR: "/mnt/data/uploads"
      CACHE_DIR: "/tmp/section_cache"
      ENABLE_TEMPORAL_WORKFLOW: "true"
      ENABLE_ASYNC_PIPELINE: "${ENABLE_ASYNC_PIPELINE:-false}"
      ENABLE_SPECULATIVE_EXECUTOR: "${ENABLE_SPECULATIVE_EXECUTOR:-false}"
      ENABLE_PROMPT_CACHING: "${ENABLE_PROMPT_CACHING:-false}"
      VECTORSTORE_BACKEND: "${VECTORSTORE_BACKEND:-faiss}"
      QDRANT_URL: "${QDRANT_URL:-http://qdrant:6333}"
      ENABLE_HYBRID_RETRIEVAL: "${ENABLE_HYBRID_RETRIEVAL:-true}"
      SEMANTIC_CACHE_ENABLED: "${SEMANTIC_CACHE_ENABLED:-true}"
      AGENTIC_INSPECTOR_WHEN_NOTES_ONLY: "${AGENTIC_INSPECTOR_WHEN_NOTES_ONLY:-false}"
      NOTES_ONLY_GENERATION: "${NOTES_ONLY_GENERATION:-true}"
      GENERATION_TIMEOUT_SECONDS: "${GENERATION_TIMEOUT_SECONDS:-3600}"
      GENERATION_STALE_SWEEP_SECONDS: "${GENERATION_STALE_SWEEP_SECONDS:-120}"
      TEMPORAL_HOST: "${TEMPORAL_HOST:-temporal:7233}"
      TEMPORAL_NAMESPACE: "${TEMPORAL_NAMESPACE:-default}"
      TEMPORAL_TASK_QUEUE: "${TEMPORAL_TASK_QUEUE:-reports}"
      MAX_CONCURRENT_LLM_CALLS: "${MAX_CONCURRENT_LLM_CALLS:-10}"
    volumes:
      - uploads:/mnt/data/uploads
      - section_cache:/tmp/section_cache
      - db_data:/data
      - faiss_index:/data/faiss_index
    depends_on:
      temporal:
        condition: service_started
    profiles:
      - temporal

volumes:
  qdrant_data:
    driver: local
  uploads:
    driver: local
  section_cache:
    driver: local
  db_data:
    driver: local
  faiss_index:
    driver: local