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

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