docker-compose.yml

# Atalaia quickstart. Two services: atalaia and a local Ollama LLM.
# Run:
#     docker compose up -d
#     # first /check call triggers a one-time ~1 GB qwen2.5:1.5b pull
#     curl -X POST -H 'content-type: text/x-diff' \
#         --data-binary @your.diff \
#         http://localhost:8080/check
#
# For real workloads, point ATALAIA_LLM_ENDPOINT/MODEL at vLLM or any
# OpenAI-compatible backend (see docs/deployment.md). This compose is
# the "try it in five minutes on a laptop" path.

services:
  ollama:
    image: ollama/ollama:latest
    container_name: atalaia-ollama
    volumes:
      - ollama-data:/root/.ollama
    healthcheck:
      test: ["CMD", "ollama", "list"]
      interval: 5s
      timeout: 5s
      retries: 12
    restart: unless-stopped

  atalaia:
    image: ghcr.io/juanfont/atalaia:latest
    container_name: atalaia
    ports:
      - "8080:8080"     # /check, /healthz, /readyz, /version
      - "9090:9090"     # /metrics
    environment:
      ATALAIA_LLM_ENDPOINT: http://ollama:11434/v1
      ATALAIA_LLM_MODEL: qwen2.5:1.5b
      # Tool calling needs --tool-call-parser configured per-model
      # on the LLM side; Ollama's coverage is uneven, so the demo
      # uses the content-parsing fallback. Flip to true once you
      # move to vLLM with a matching parser.
      ATALAIA_LLM_USE_TOOLS: "false"
    depends_on:
      ollama:
        condition: service_healthy
    restart: unless-stopped

volumes:
  ollama-data: