docker-compose.yml

services:
  neo4j:
    image: neo4j:5.15-community
    container_name: neo4j_db
    ports:
      - "7474:7474"  # HTTP Browser UI
      - "7687:7687"  # Bolt protocol
    volumes:
      - ./data/neo4j:/data
      - ./logs/neo4j:/logs
      - ./data/neo4j/import:/var/lib/neo4j/import
    environment:
      - NEO4J_AUTH=neo4j/${NEO4J_PASSWORD:-password}
      - NEO4J_dbms_memory_pagecache_size=512M
      - NEO4J_dbms_memory_heap_initial__size=512M
      - NEO4J_dbms_memory_heap_max__size=2G
      - NEO4J_PLUGINS=["apoc"]
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "cypher-shell -u neo4j -p $${NEO4J_PASSWORD:-password} 'RETURN 1' || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    networks:
      - mcp_network

  copilot-api:
    image: timothyswt/copilot-api:latest
    container_name: copilot_api_server
    ports:
      - "4141:4141"  # Fixed: copilot-api listens on 4141, not 3000
    volumes:
      - ./copilot-data:/root/.local/share/copilot-api  # Persist GitHub token
    environment:
      - NODE_ENV=production
      # Remove PORT=3000, the app uses 4141 by default
    restart: unless-stopped
    healthcheck:
      # Use CMD-SHELL so shell operators (||) work and allow a proper HTTP probe
      test: ["CMD-SHELL", "wget --spider -q http://localhost:4141/ || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 15s
    networks:
      - mcp_network

# Ollama (Legacy - kept for reference, can be removed)
  # ollama:
  #   build:
  #     context: ./docker/ollama
  #     dockerfile: Dockerfile
  #     args:
  #       - EMBEDDING_MODEL=${MIMIR_EMBEDDINGS_MODEL:-bge-m3}
  #     tags:
  #       - mimir-ollama:${VERSION:-1.0.0}
  #       - mimir-ollama:latest
  #   image: mimir-ollama:${VERSION:-1.0.0}
  #   container_name: ollama_server
  #   ports:
  #     - "11434:11434"
  #   volumes:
  #     - type: bind
  #       source: ollama_models
  #       target: /root/.ollama
  #   environment:
  #     - OLLAMA_HOST=0.0.0.0:11434
  #     - OLLAMA_ORIGINS=*
  #   restart: unless-stopped
  #   healthcheck:
  #     test: ["CMD", "ollama", "list"]
  #     interval: 10s
  #     timeout: 5s
  #     retries: 5
  #     start_period: 30s
  #   networks:
  #     - mcp_network
  #   deploy:
  #     resources:
  #       reservations:
  #         devices:
  #           - driver: nvidia
  #             count: 1
  #             capabilities: [gpu]

  # llama.cpp server - OpenAI-compatible embeddings
  llama-server:
    image: ghcr.io/ggml-org/llama.cpp:server-cuda
    container_name: llama_server
    ports:
      - "11434:8080"  # External 11434 -> Internal 8080 (llama.cpp default)
    volumes:
      - type: bind
        source: ollama_models
        target: /models
        read_only: true
    environment:
      # Model Configuration - using bge-m3
      - LLAMA_ARG_MODEL=/models/models/blobs/sha256-819c2adf5ce6df2b6bd2ae4ca90d2a69f060afeb438d0c171db57daa02e39c3d
      - LLAMA_ARG_ALIAS=bge-m3
      
      # Server Configuration
      - LLAMA_ARG_HOST=0.0.0.0
      - LLAMA_ARG_PORT=8080
      - LLAMA_ARG_CTX_SIZE=2048
      - LLAMA_ARG_N_PARALLEL=4
      
      # Embeddings-specific
      - LLAMA_ARG_EMBEDDINGS=true
      - LLAMA_ARG_POOLING=mean
      
      # Performance
      - LLAMA_ARG_THREADS=-1
      - LLAMA_ARG_NO_MMAP=false
      
      # GPU support
      - LLAMA_ARG_N_GPU_LAYERS=99
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    networks:
      - mcp_network
    # Uncomment if you have GPU support (NVIDIA)
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  # Vision-Language server for image description (qwen2.5-vl via llama.cpp)
  # Generates text descriptions of images, which are then embedded by text model
  # Uncomment to enable image indexing with MIMIR_EMBEDDINGS_IMAGES_DESCRIBE_MODE=true
  # To use: 1) Uncomment service, 2) docker-compose up -d llama-vl-server
  # llama-vl-server:
  #   image: timothyswt/llama-cpp-server-arm64-qwen2.5-vl:7b  # or :2b for lighter model
  #   container_name: llama_vl_server
  #   ports:
  #     - "8081:8080"  # Different external port to avoid conflict with llama-server
  #   environment:
  #     # Model is baked into image, these are runtime overrides
  #     # Context size: 128K tokens (7b/72b), 32K tokens (2b)
  #     - LLAMA_ARG_CTX_SIZE=131072  # 128K tokens for 7b model (use 32768 for 2b)
  #     - LLAMA_ARG_N_PARALLEL=4
  #     - LLAMA_ARG_THREADS=-1  # Use all available threads
  #     - LLAMA_ARG_HOST=0.0.0.0
  #     - LLAMA_ARG_PORT=8080
  #     # Vision-specific settings
  #     - LLAMA_ARG_TEMPERATURE=0.7
  #     - LLAMA_ARG_TOP_K=20
  #     - LLAMA_ARG_TOP_P=0.95
  #   restart: unless-stopped
  #   healthcheck:
  #     test: ["CMD", "curl", "-f", "http://localhost:8080/health"]
  #     interval: 30s
  #     timeout: 10s
  #     retries: 3
  #     start_period: 60s  # VL models take longer to load
  #   networks:
  #     - mcp_network
  #   # Uncomment for GPU support (NVIDIA)
  #   # deploy:
  #   #   resources:
  #   #     reservations:
  #   #       devices:
  #   #         - driver: nvidia
  #   #           count: 1
  #   #           capabilities: [gpu]

  mimir-server:
    build:
      context: .
      dockerfile: Dockerfile
      tags:
        - timothyswt/mimir-server:${VERSION:-1.0.0}
        - timothyswt/mimir-server:latest
    image: timothyswt/mimir-server:latest
    container_name: mimir_server
    restart: unless-stopped
    environment:
      # Database Configuration
      - NEO4J_URI=bolt://neo4j:7687
      - NEO4J_USER=neo4j
      - NEO4J_PASSWORD=${NEO4J_PASSWORD:-password}
      
      # Server Configuration
      - NODE_ENV=production
      - PORT=3000
      # - NODE_TLS_REJECT_UNAUTHORIZED=${NODE_TLS_REJECT_UNAUTHORIZED:-1}
      
      # Workspace Configuration
      - WORKSPACE_ROOT=/workspace
      - HOST_WORKSPACE_ROOT=${HOST_WORKSPACE_ROOT:-~/src}  # Pass through from host
      - HOST_HOME=${HOME}  # Host's home directory for expanding ~ in HOST_WORKSPACE_ROOT
      
      # LLM API Configuration (Copilot - OpenAI-compatible)
      - MIMIR_DEFAULT_PROVIDER=${MIMIR_DEFAULT_PROVIDER:-copilot}
      - MIMIR_LLM_API=${MIMIR_LLM_API:-http://copilot-api:4141/v1}
      - MIMIR_LLM_API_PATH=${MIMIR_LLM_API_PATH:-/chat/completions}
      - MIMIR_LLM_API_MODELS_PATH=${MIMIR_LLM_API_MODELS_PATH:-/models}
      - MIMIR_LLM_API_KEY=${MIMIR_LLM_API_KEY:-dummy-key-for-proxy}
      
      # PCTX Integration (Code Mode for 90-98% token reduction)
      - PCTX_URL=${PCTX_URL:-http://host.docker.internal:8080}
      - PCTX_ENABLED=${PCTX_ENABLED:-false}
      
      # Provider and Model Configuration (100% dynamic - no config file needed)
      - MIMIR_DEFAULT_MODEL=${MIMIR_DEFAULT_MODEL:-gpt-4.1}
      
      # Per-Agent Model Configuration (optional overrides)
      - MIMIR_PM_MODEL=${MIMIR_PM_MODEL:-gpt-4.1}
      - MIMIR_WORKER_MODEL=${MIMIR_WORKER_MODEL:-gpt-4.1}
      - MIMIR_QC_MODEL=${MIMIR_QC_MODEL:-gpt-4.1}
      
      # Context Window Configuration
      - MIMIR_DEFAULT_CONTEXT_WINDOW=${MIMIR_DEFAULT_CONTEXT_WINDOW:-128000}
      
      # Embeddings API Configuration (llama.cpp server - OpenAI-compatible)
      - MIMIR_EMBEDDINGS_PROVIDER=${MIMIR_EMBEDDINGS_PROVIDER:-openai}
      - MIMIR_EMBEDDINGS_API=${MIMIR_EMBEDDINGS_API:-http://llama-server:8080}
      - MIMIR_EMBEDDINGS_API_PATH=${MIMIR_EMBEDDINGS_API_PATH:-/v1/embeddings}
      - MIMIR_EMBEDDINGS_API_MODELS_PATH=${MIMIR_EMBEDDINGS_API_MODELS_PATH:-/v1/models}
      - MIMIR_EMBEDDINGS_API_KEY=${MIMIR_EMBEDDINGS_API_KEY:-dummy-key}
      # Embeddings Configuration (llama.cpp uses OpenAI-compatible format)
      - MIMIR_EMBEDDINGS_ENABLED=${MIMIR_EMBEDDINGS_ENABLED:-true}
      - MIMIR_EMBEDDINGS_MODEL=${MIMIR_EMBEDDINGS_MODEL:-bge-m3}
      - MIMIR_EMBEDDINGS_DIMENSIONS=${MIMIR_EMBEDDINGS_DIMENSIONS:-1024}
      - MIMIR_EMBEDDINGS_CHUNK_SIZE=${MIMIR_EMBEDDINGS_CHUNK_SIZE:-768}
      - MIMIR_EMBEDDINGS_CHUNK_OVERLAP=${MIMIR_EMBEDDINGS_CHUNK_OVERLAP:-10}
      - MIMIR_EMBEDDINGS_DELAY_MS=${MIMIR_EMBEDDINGS_DELAY_MS:-100}
      - MIMIR_EMBEDDINGS_MAX_RETRIES=${MIMIR_EMBEDDINGS_MAX_RETRIES:-3}
      # Image/VL Embeddings Configuration (Vision-Language models)
      # Image Embeddings Control (disabled by default for safety)
      - MIMIR_EMBEDDINGS_IMAGES=${MIMIR_EMBEDDINGS_IMAGES:-false}  # Default: disabled
      - MIMIR_EMBEDDINGS_IMAGES_DESCRIBE_MODE=${MIMIR_EMBEDDINGS_IMAGES_DESCRIBE_MODE:-true}  # Default: VL description mode
      - MIMIR_EMBEDDINGS_VL_PROVIDER=${MIMIR_EMBEDDINGS_VL_PROVIDER:-}
      - MIMIR_EMBEDDINGS_VL_API=${MIMIR_EMBEDDINGS_VL_API:-}
      - MIMIR_EMBEDDINGS_VL_API_PATH=${MIMIR_EMBEDDINGS_VL_API_PATH:-}
      - MIMIR_EMBEDDINGS_VL_API_KEY=${MIMIR_EMBEDDINGS_VL_API_KEY:-}
      - MIMIR_EMBEDDINGS_VL_MODEL=${MIMIR_EMBEDDINGS_VL_MODEL:-}
      - MIMIR_EMBEDDINGS_VL_DIMENSIONS=${MIMIR_EMBEDDINGS_VL_DIMENSIONS:-}
      
      # Indexing Configuration
      - MIMIR_INDEXING_THREADS=${MIMIR_INDEXING_THREADS:-1}
      
      # Feature Flags
      - MIMIR_FEATURE_PM_MODEL_SUGGESTIONS=${MIMIR_FEATURE_PM_MODEL_SUGGESTIONS:-true}
      - MIMIR_AUTO_INDEX_DOCS=${MIMIR_AUTO_INDEX_DOCS:-true}
      
      # Agent Execution Limits
      - MIMIR_AGENT_RECURSION_LIMIT=${MIMIR_AGENT_RECURSION_LIMIT:-100}
      
      # Security Configuration
      - MIMIR_ENABLE_SECURITY=${MIMIR_ENABLE_SECURITY:-false}
      - MIMIR_DEV_USER_ADMIN=${MIMIR_DEV_USER_ADMIN}
      - MIMIR_DEV_USER_DEVELOPER=${MIMIR_DEV_USER_DEVELOPER}
      - MIMIR_DEV_USER_ANALYST=${MIMIR_DEV_USER_ANALYST}
      - MIMIR_DEV_USER_VIEWER=${MIMIR_DEV_USER_VIEWER}
      - MIMIR_JWT_SECRET=${MIMIR_JWT_SECRET}

      # OAuth Configuration (explicit endpoint URLs - provider-specific)
      - MIMIR_AUTH_PROVIDER=${MIMIR_AUTH_PROVIDER}
      - MIMIR_OAUTH_AUTHORIZATION_URL=${MIMIR_OAUTH_AUTHORIZATION_URL}  # Full authorization endpoint URL
      - MIMIR_OAUTH_TOKEN_URL=${MIMIR_OAUTH_TOKEN_URL}                  # Full token endpoint URL
      - MIMIR_OAUTH_USERINFO_URL=${MIMIR_OAUTH_USERINFO_URL}            # Full userinfo endpoint URL
      - MIMIR_OAUTH_CLIENT_ID=${MIMIR_OAUTH_CLIENT_ID}
      - MIMIR_OAUTH_CLIENT_SECRET=${MIMIR_OAUTH_CLIENT_SECRET}
      - MIMIR_OAUTH_CALLBACK_URL=${MIMIR_OAUTH_CALLBACK_URL}
      - MIMIR_OAUTH_ALLOW_HTTP=${MIMIR_OAUTH_ALLOW_HTTP}
      
      # Advanced Configuration
      - MIMIR_PARALLEL_EXECUTION=${MIMIR_PARALLEL_EXECUTION:-false}
      - MIMIR_INSTALL_DIR=${MIMIR_INSTALL_DIR:-/app}
      - MIMIR_AGENTS_DIR=${MIMIR_AGENTS_DIR:-/app/docs/agents}
    volumes:
      - ./data:/app/data
      - ./logs:/app/logs
      - ${HOST_WORKSPACE_ROOT:-~/src}:${WORKSPACE_ROOT:-/workspace}  # Read-write for file editing (remove :ro)
      # - ${HOST_WORKSPACE_ROOT:-~/src}:/workspace:ro  # Read-write for file editing (remove :ro)
    ports:
      - "9042:3000"
    healthcheck:
      test: ["CMD", "node", "-e", "require('http').get('http://localhost:3000/health', (res) => process.exit(res.statusCode === 200 ? 0 : 1)).on('error', () => process.exit(1))"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    depends_on:
      neo4j:
        condition: service_healthy
      copilot-api:
        condition: service_healthy
      # Ollama is required for embeddings but you can use a local server
      llama-server:
        condition: service_healthy
      # llama-vl-server:
      #   condition: service_healthy
    networks:
      - mcp_network

  # # Open-WebUI with Mimir Pipeline Integration
  # open-webui:
  #   build:
  #     context: .
  #     dockerfile: docker/open-webui/Dockerfile
  #     tags:
  #       - mimir-open-webui:${VERSION:-1.0.0}
  #       - mimir-open-webui:latest
  #   image: mimir-open-webui:${VERSION:-1.0.0}
  #   container_name: mimir-open-webui
  #   ports:
  #     - "3000:8080"
  #   volumes:
  #     - open-webui:/app/backend/data
  #   environment:
  #     # OpenAI-Compatible API Configuration (Copilot API)
  #     - OPENAI_API_BASE_URL=http://copilot-api:4141/v1
  #     # - OPENAI_API_BASE_URL=http://host.docker.internal:4141/v1
  #     - OPENAI_API_KEY=sk-copilot-dummy  # Dummy key for copilot-api
      
  #     # Disable Ollama (using Copilot API instead)
  #     - ENABLE_OLLAMA_API=false
  #     - OLLAMA_BASE_URL=
      
  #     # WebUI Configuration
  #     - WEBUI_NAME=Mimir Multi-Agent Orchestrator
  #     - WEBUI_URL=http://localhost:3000
  #     - DEFAULT_MODELS=gpt-4.1  # Default selected model
      
  #     # Enable OpenAI API
  #     - ENABLE_OPENAI_API=true
  #   depends_on:
  #     - mcp-server
  #     - copilot-api
  #   networks:
  #     - mcp_network
  #   restart: unless-stopped
  #   extra_hosts:
  #     - "host.docker.internal:host-gateway"  # Allow access to host's localhost

volumes:
  open-webui:
  ollama_models:

networks:
  mcp_network:
    driver: bridge