docker-compose-bindmount-checkout.yml

services:
  qdrant:
    image: qdrant/qdrant:latest
    container_name: qdrant-db
    # Expose Qdrant database APIs to the host
    # 6333 = HTTP API, 6334 = gRPC
    ports:
      - "6333:6333"
      - "6334:6334"
    volumes:
      - qdrant_storage:/qdrant/storage

  mcp:
    build:
      context: .
      dockerfile: Dockerfile.mcp
    container_name: mcp-search
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      # Ensure the server binds to all interfaces for container networking
      - FASTMCP_HOST=${FASTMCP_HOST}
      - FASTMCP_PORT=${FASTMCP_PORT}
      - QDRANT_URL=${QDRANT_URL}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - EMBEDDING_PROVIDER=${EMBEDDING_PROVIDER}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
      - TOOL_STORE_DESCRIPTION=${TOOL_STORE_DESCRIPTION}
      - TOOL_FIND_DESCRIPTION=${TOOL_FIND_DESCRIPTION}
      - FASTMCP_HEALTH_PORT=18000

    # SSE endpoint for IDE agents at http://localhost:8000/sse
    ports:
      - "18000:18000"

      - "8000:8000"

    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:ro

  mcp_indexer:
    build:
      context: .
      dockerfile: Dockerfile.mcp-indexer
    container_name: mcp-indexer
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - FASTMCP_HEALTH_PORT=18001
      - FASTMCP_HOST=${FASTMCP_HOST}
      - FASTMCP_INDEXER_PORT=${FASTMCP_INDEXER_PORT}
      - QDRANT_URL=${QDRANT_URL}
      - DEBUG_CONTEXT_ANSWER=${DEBUG_CONTEXT_ANSWER:-1}
      - REFRAG_DECODER=${REFRAG_DECODER:-1}
      - LLAMACPP_URL=${LLAMACPP_URL:-http://llamacpp:8080}
      - USE_GPU_DECODER=${USE_GPU_DECODER:-0}
      - LLAMACPP_TIMEOUT_SEC=${LLAMACPP_TIMEOUT_SEC:-180}
      - CTX_REQUIRE_IDENTIFIER=${CTX_REQUIRE_IDENTIFIER:-0}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
    # SSE endpoint for IDE agents at http://localhost:${FASTMCP_INDEXER_PORT:-8001}/sse
    ports:
      - "${FASTMCP_INDEXER_PORT:-8001}:8001"
      - "18001:18001"
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work


  mcp_http:
    build:
      context: .
      dockerfile: Dockerfile.mcp
    container_name: mcp-search-http
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - FASTMCP_HOST=${FASTMCP_HOST}
      - FASTMCP_PORT=8000
      - FASTMCP_TRANSPORT=${FASTMCP_HTTP_TRANSPORT}
      - QDRANT_URL=${QDRANT_URL}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - EMBEDDING_PROVIDER=${EMBEDDING_PROVIDER}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
      - TOOL_STORE_DESCRIPTION=${TOOL_STORE_DESCRIPTION}
      - TOOL_FIND_DESCRIPTION=${TOOL_FIND_DESCRIPTION}
      - FASTMCP_HEALTH_PORT=18000
    # Streamable HTTP endpoint for IDE agents at http://localhost:${FASTMCP_HTTP_PORT:-8002}/mcp/
    ports:
      - "${FASTMCP_HTTP_HEALTH_PORT:-18002}:18000"
      - "${FASTMCP_HTTP_PORT:-8002}:8000"
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:ro

  mcp_indexer_http:
    build:
      context: .
      dockerfile: Dockerfile.mcp-indexer
    container_name: mcp-indexer-http
    depends_on:
      - qdrant
      - llamacpp
    env_file:
      - .env
    environment:
      - FASTMCP_HOST=${FASTMCP_HOST}
      - FASTMCP_INDEXER_PORT=8001
      - FASTMCP_TRANSPORT=${FASTMCP_HTTP_TRANSPORT}
      - QDRANT_URL=${QDRANT_URL}
      - FASTMCP_HEALTH_PORT=18001
      - DEBUG_CONTEXT_ANSWER=${DEBUG_CONTEXT_ANSWER:-1}
      - REFRAG_DECODER=${REFRAG_DECODER:-1}
      - LLAMACPP_URL=${LLAMACPP_URL:-http://llamacpp:8080}
      - USE_GPU_DECODER=${USE_GPU_DECODER:-0}
      - LLAMACPP_TIMEOUT_SEC=${LLAMACPP_TIMEOUT_SEC:-180}
      - CTX_REQUIRE_IDENTIFIER=${CTX_REQUIRE_IDENTIFIER:-0}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
    # Streamable HTTP endpoint for IDE agents at http://localhost:${FASTMCP_INDEXER_HTTP_PORT:-8003}/mcp/
    ports:
      - "${FASTMCP_INDEXER_HTTP_PORT:-8003}:8001"
      - "${FASTMCP_INDEXER_HTTP_HEALTH_PORT:-18003}:18001"
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work

  llamacpp:
    image: ghcr.io/ggml-org/llama.cpp:server
    container_name: llama-decoder
    # Optional sidecar providing a text-generation API on :8080
    # No behavior change unless REFRAG_DECODER=1
    environment:
      - LLAMACPP_CTX_SIZE=${LLAMACPP_CTX_SIZE:-8192}
      - LLAMACPP_HOST=0.0.0.0
      - LLAMACPP_PORT=8080
      - LLAMACPP_USE_GPU=${LLAMACPP_USE_GPU:-0}
      - LLAMACPP_GPU_LAYERS=${LLAMACPP_GPU_LAYERS:-0}
      - LLAMACPP_GPU_SPLIT=${LLAMACPP_GPU_SPLIT:-}
      - LLAMACPP_THREADS=${LLAMACPP_THREADS:-}
      - LLAMACPP_EXTRA_ARGS=${LLAMACPP_EXTRA_ARGS:-}
      - LLAMACPP_NO_WARMUP=${LLAMACPP_NO_WARMUP:-1}
      - LLAMACPP_TEMPERATURE=${LLAMACPP_TEMPERATURE:-}
    ports:
      - "8080:8080"
    volumes:
      - ./models:/models:ro
    entrypoint: ["/bin/sh","-lc"]
    command:
      - |
        set -e
        ARGS="--model /models/model.gguf --host ${LLAMACPP_HOST:-0.0.0.0} --port ${LLAMACPP_PORT:-8080} --ctx-size ${LLAMACPP_CTX_SIZE:-8192}"
        if [ "${LLAMACPP_USE_GPU:-0}" = "1" ]; then
          LAYERS="${LLAMACPP_GPU_LAYERS:--1}"
        else
          LAYERS="${LLAMACPP_GPU_LAYERS:-0}"
        fi
        ARGS="$$ARGS --n-gpu-layers $${LAYERS}"
        if [ -n "${LLAMACPP_GPU_SPLIT:-}" ]; then
          ARGS="$$ARGS --tensor-split ${LLAMACPP_GPU_SPLIT}"
        fi
        if [ "${LLAMACPP_NO_WARMUP:-1}" != "0" ]; then
          ARGS="$$ARGS --no-warmup"
        fi
        if [ -n "${LLAMACPP_THREADS:-}" ]; then
          ARGS="$$ARGS --threads ${LLAMACPP_THREADS}"
        fi
        if [ -n "${LLAMACPP_EXTRA_ARGS:-}" ]; then
          ARGS="$$ARGS ${LLAMACPP_EXTRA_ARGS}"
        fi
        exec /app/llama-server $$ARGS

  indexer:
    build:
      context: .
      dockerfile: Dockerfile.indexer
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - QDRANT_URL=${QDRANT_URL}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - HF_HOME=/tmp/huggingface
      - HF_HUB_CACHE=/tmp/huggingface/hub
      - HUGGINGFACE_HUB_CACHE=/tmp/huggingface/hub
      - TRANSFORMERS_CACHE=/tmp/huggingface/transformers
      - FASTEMBED_CACHE_PATH=/tmp/huggingface/fastembed
      - HF_HUB_DISABLE_XET=1
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
    working_dir: /work
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:ro
      - ${HOST_INDEX_PATH:-.}/.codebase:/work/.codebase:rw

    entrypoint: ["sh", "-c", "mkdir -p /tmp/huggingface/hub /tmp/huggingface/transformers /tmp/huggingface/fastembed && exec python /app/scripts/ingest_code.py"]

  watcher:
    build:
      context: .
      dockerfile: Dockerfile.indexer
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - QDRANT_URL=${QDRANT_URL}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - HF_HOME=/tmp/huggingface
      - HF_HUB_CACHE=/tmp/huggingface/hub
      - HUGGINGFACE_HUB_CACHE=/tmp/huggingface/hub
      - TRANSFORMERS_CACHE=/tmp/huggingface/transformers
      - FASTEMBED_CACHE_PATH=/tmp/huggingface/fastembed
      - HF_HUB_DISABLE_XET=1
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
      - WATCH_ROOT=/work
      # Watcher-specific backpressure & timeouts (safer defaults)
      - QDRANT_TIMEOUT=60
      - MAX_MICRO_CHUNKS_PER_FILE=${MAX_MICRO_CHUNKS_PER_FILE:-200}
      - INDEX_UPSERT_BATCH=128
      - INDEX_UPSERT_RETRIES=5
      - WATCH_DEBOUNCE_SECS=${WATCH_DEBOUNCE_SECS:-1.5}
    working_dir: /work
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:ro
      - ${HOST_INDEX_PATH:-.}/.codebase:/work/.codebase:rw
    entrypoint: ["sh", "-c", "mkdir -p /tmp/huggingface/hub /tmp/huggingface/transformers /tmp/huggingface/fastembed && exec python /app/scripts/watch_index.py"]


  upload_service:
    build:
      context: .
      dockerfile: Dockerfile.upload-service
    container_name: upload-service
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - UPLOAD_SERVICE_HOST=0.0.0.0
      - UPLOAD_SERVICE_PORT=8002
      - QDRANT_URL=${QDRANT_URL}
      - WORK_DIR=/work
      - CTXCE_ADMIN_COLLECTION_DELETE_ENABLED=0
      - CTXCE_COLLECTION_REGISTRY_UNDELETE_ON_DISCOVERY=${CTXCE_COLLECTION_REGISTRY_UNDELETE_ON_DISCOVERY:-0}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
      - EMBEDDING_MODEL=${EMBEDDING_MODEL}
      - EMBEDDING_PROVIDER=${EMBEDDING_PROVIDER}
      - QWEN3_EMBEDDING_ENABLED=${QWEN3_EMBEDDING_ENABLED:-0}
      - QWEN3_QUERY_INSTRUCTION=${QWEN3_QUERY_INSTRUCTION:-1}
      - QWEN3_INSTRUCTION_TEXT=${QWEN3_INSTRUCTION_TEXT}
    ports:
      - "8004:8002"
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:rw
      - ${HOST_INDEX_PATH:-.}/.codebase:/work/.codebase:rw
    user: "0:0"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8002/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    restart: unless-stopped

  init_payload:
    build:
      context: .
      dockerfile: Dockerfile.indexer
    depends_on:
      - qdrant
    env_file:
      - .env
    environment:
      - QDRANT_URL=${QDRANT_URL}
      - COLLECTION_NAME=${COLLECTION_NAME:-codebase}
    working_dir: /work
    volumes:
      - ${HOST_INDEX_PATH:-.}:/work:ro
      - ${HOST_INDEX_PATH:-.}/.codebase:/work/.codebase:rw

    entrypoint: ["python", "/app/scripts/create_indexes.py"]

volumes:
  qdrant_storage:
    driver: local