M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

services: nornicdb: build: context: ./nornicdb dockerfile: Dockerfile.cuda tags: - timothyswt/nornicdb:cuda image: timothyswt/nornicdb:cuda container_name: nornicdb ports: - "7474:7474" # HTTP API & Browser UI - "7687:7687" # Bolt protocol (Neo4j compatible) volumes: - ./data/nornicdb:/data - ./data/nornicdb/export:/data/export # For loading Mimir exports environment: - NORNICDB_DATA_DIR=/data - NORNICDB_HTTP_PORT=7474 - NORNICDB_BOLT_PORT=7687 # - NORNICDB_EMBEDDING_URL=http://llama-server:8080 # - NORNICDB_EMBEDDING_MODEL=bge-m3 # - NORNICDB_EMBEDDING_DIM=1024 # # Authentication (disabled by default for dev) - NORNICDB_NO_AUTH=${NORNICDB_NO_AUTH:-true} - NORNICDB_ADMIN_PASSWORD=${NEO4J_PASSWORD:-password} # Load export data on startup (disabled - starting fresh) - NORNICDB_LOAD_EXPORT= # GPU Configuration (auto-detected, can override) - NORNICDB_GPU_ENABLED=${NORNICDB_GPU_ENABLED:-true} - NORNICDB_GPU_DEVICE_ID=${NORNICDB_GPU_DEVICE_ID:-0} restart: unless-stopped healthcheck: test: ["CMD-SHELL", "wget --spider -q http://localhost:7474/health || exit 1"] interval: 30s timeout: 10s retries: 3 start_period: 10s depends_on: llama-server: condition: service_healthy networks: - mcp_network deploy: resources: reservations: devices: - driver: nvidia count: 1 capabilities: [gpu, compute, utility] 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: timothyswt/llama-cpp-server-amd64-mxbai-cuda:latest container_name: llama_server ports: - "11434:8080" # External 11434 -> Internal 8080 (llama.cpp default) # Note: Model is bundled in the image (bge-m3, 1024 dims) # No volumes needed unless you want to add additional models restart: unless-stopped healthcheck: test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"] interval: 30s timeout: 10s retries: 3 start_period: 30s networks: - mcp_network deploy: resources: reservations: devices: - driver: nvidia count: 1 capabilities: [gpu] # llama.cpp Vision Server - 2B model (faster, lower memory) # Qwen2.5-VL 2B with vision understanding and CUDA acceleration # Build with: npm run llama:build-qwen-2b-cuda # For use without cuda `image: timothyswt/llama-cpp-server-amd64-qwen-vl-2b:latest` # # llama-vl-server-2b: # image: timothyswt/llama-cpp-server-amd64-qwen-vl-2b-cuda:latest # container_name: llama_server_vision_2b # ports: # - "8081:8080" # Vision 2B API endpoint # restart: unless-stopped # healthcheck: # test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"] # interval: 30s # timeout: 10s # retries: 3 # start_period: 60s # Vision models take longer to load # networks: # - mcp_network # deploy: # resources: # limits: # memory: 8G # 2B model + vision projector # reservations: # memory: 4G # devices: # - driver: nvidia # count: 1 # capabilities: [gpu] # llama.cpp Vision Server - 7B model (higher quality, more memory) # Qwen2.5-VL 7B with vision understanding and CUDA acceleration # Build with: npm run llama:build-qwen-7b-cuda # For use without cuda `image: timothyswt/llama-cpp-server-amd64-qwen-vl-7b:latest` # # llama-vl-server-7b: # image: timothyswt/llama-cpp-server-amd64-qwen-vl-7b-cuda:latest # container_name: llama_server_vision_7b # ports: # - "8082:8080" # Vision 7B API endpoint # restart: unless-stopped # healthcheck: # test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"] # interval: 30s # timeout: 10s # retries: 3 # start_period: 90s # Larger models take even longer # networks: # - mcp_network # deploy: # resources: # limits: # memory: 16G # 7B models need more RAM # reservations: # memory: 8G # devices: # - driver: nvidia # count: 1 # capabilities: [gpu] mimir-server: build: context: . dockerfile: Dockerfile tags: - mimir-server:${VERSION:-1.0.0} - mimir-server:latest image: mimir-server:${VERSION:-1.0.0} container_name: mimir_server restart: unless-stopped environment: # Database Configuration - NEO4J_URI=bolt://nornicdb:7687 - NEO4J_USER=admin - NEO4J_PASSWORD=${NEO4J_PASSWORD:-admin} # 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} # Pass through from host - HOST_HOME=${HOME} # Host's home directory for expanding ~ in HOST_WORKSPACE_ROOT # LLM API Configuration - MIMIR_DEFAULT_PROVIDER=${MIMIR_DEFAULT_PROVIDER:-copilot} - MIMIR_LLM_API=${MIMIR_LLM_API:-http://copilot-api:4141} - MIMIR_LLM_API_PATH=${MIMIR_LLM_API_PATH:-/v1/chat/completions} - MIMIR_LLM_API_MODELS_PATH=${MIMIR_LLM_API_MODELS_PATH:-/v1/models} - MIMIR_LLM_API_KEY=${MIMIR_LLM_API_KEY:-dummy-key} # 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:-} - MIMIR_WORKER_MODEL=${MIMIR_WORKER_MODEL:-} - MIMIR_QC_MODEL=${MIMIR_QC_MODEL:-} # Context Window Configuration - MIMIR_DEFAULT_CONTEXT_WINDOW=${MIMIR_DEFAULT_CONTEXT_WINDOW:-128000} # Embeddings API Configuration - MIMIR_EMBEDDINGS_PROVIDER=${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 - 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=${MIMIR_EMBEDDINGS_CHUNK_OVERLAP:-100} - MIMIR_EMBEDDINGS_DELAY_MS=${MIMIR_EMBEDDINGS_DELAY_MS:-0} - MIMIR_EMBEDDINGS_MAX_RETRIES=${MIMIR_EMBEDDINGS_MAX_RETRIES:-3} # Image Embeddings Control (disabled by default for safety) - MIMIR_EMBEDDINGS_IMAGES=${MIMIR_EMBEDDINGS_IMAGES:-false} # Default: disabled - MIMIR_EMBEDDINGS_IMAGES_DESCRIBE_MODE=${MIMIR_EMBEDDINGS_IMAGES_DESCRIBE_MODE:-true} # Default: VL description mode # Qwen2.5-VL Configuration (for llama.cpp server) - MIMIR_EMBEDDINGS_VL_PROVIDER=${MIMIR_EMBEDDINGS_VL_PROVIDER:-llama.cpp} - MIMIR_EMBEDDINGS_VL_API=${MIMIR_EMBEDDINGS_VL_API:-http://llama-vl-server-2b:8080} - MIMIR_EMBEDDINGS_VL_API_PATH=${MIMIR_EMBEDDINGS_VL_API_PATH:-/v1/chat/completions} - MIMIR_EMBEDDINGS_VL_API_KEY=${MIMIR_EMBEDDINGS_VL_API_KEY:-dummy-key} - MIMIR_EMBEDDINGS_VL_MODEL=${MIMIR_EMBEDDINGS_VL_MODEL:-qwen2.5-vl} - MIMIR_EMBEDDINGS_VL_CONTEXT_SIZE=${MIMIR_EMBEDDINGS_VL_CONTEXT_SIZE:-131072} # 128K tokens (7b/72b) - MIMIR_EMBEDDINGS_VL_MAX_TOKENS=${MIMIR_EMBEDDINGS_VL_MAX_TOKENS:-2048} # Max description length - MIMIR_EMBEDDINGS_VL_TEMPERATURE=${MIMIR_EMBEDDINGS_VL_TEMPERATURE:-0.7} - MIMIR_EMBEDDINGS_VL_DIMENSIONS=${MIMIR_EMBEDDINGS_VL_DIMENSIONS:-768} # Falls back to text model dims - MIMIR_EMBEDDINGS_VL_TIMEOUT=${MIMIR_EMBEDDINGS_VL_TIMEOUT:-180000} # 3 minutes (VL processing is slow) # Indexing Configuration - MIMIR_INDEXING_THREADS=${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} # 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} 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: nornicdb: condition: service_healthy copilot-api: condition: service_healthy # llama.cpp server for embeddings llama-server: condition: service_healthy # llama.cpp server for image embeddings must set # llama-vl-server: # condition: service_healthy networks: - mcp_network networks: mcp_network: driver: bridge