Skip to content

AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.

18.0 KB
Raw Blame
# Environment Variable Loading Order (first → last, later overrides earlier):
# 1. backend/.env.default - Default values for all settings
# 2. backend/.env - User's custom configuration (if exists)
# 3. environment key - Docker-specific overrides defined below
# 4. Shell environment - Variables exported before running docker compose
# 5. CLI arguments - docker compose run -e VAR=value

# Common backend environment - Docker service names
x-backend-env:
  &backend-env # Docker internal service hostnames (override localhost defaults)
  PYRO_HOST: "0.0.0.0"
  AGENTSERVER_HOST: rest_server
  SCHEDULER_HOST: scheduler_server
  DATABASEMANAGER_HOST: database_manager
  EXECUTIONMANAGER_HOST: executor
  NOTIFICATIONMANAGER_HOST: notification_server
  PLATFORMLINKINGMANAGER_HOST: platform_linking_manager
  CLAMAV_SERVICE_HOST: clamav
  DB_HOST: db
  REDIS_HOST: redis-0
  REDIS_PORT: "17000"
  # The 3-container local cluster announces distinct compose hostnames per
  # shard (redis-0/1/2), each resolvable via Docker DNS inside the compose
  # network. Bypass the HOST-pinning address_remap so CLUSTER SLOTS uses the
  # announced addresses directly — see ``backend.data.redis_client``.
  REDIS_USE_ANNOUNCED_ADDRESS: "true"
  RABBITMQ_HOST: rabbitmq
  GRAPHITI_FALKORDB_HOST: falkordb
  GRAPHITI_FALKORDB_PORT: "6379"
  # JWKS endpoint of the Better Auth service embedded in the frontend.
  # Containers reach it via the compose service name, not localhost.
  JWT_JWKS_URL: http://frontend:3000/api/auth/jwks
  # Database connection string for Docker network
  # This cannot be constructed like in .env because we cannot interpolate values set here (DB_HOST)
  DATABASE_URL: postgresql://postgres:your-super-secret-and-long-postgres-password@db:5432/postgres?connect_timeout=60&schema=platform
  DIRECT_URL: postgresql://postgres:your-super-secret-and-long-postgres-password@db:5432/postgres?connect_timeout=60&schema=platform
  FRONTEND_BASE_URL: ${FRONTEND_BASE_URL:-http://localhost:3000}

# Common env_file configuration for backend services
x-backend-env-files: &backend-env-files
  env_file:
    - backend/.env.default # Base defaults (always exists)
    - path: backend/.env # User overrides (optional)
      required: false

x-codex-tmpfs: &codex-tmpfs
  - /run/autogpt-codex:rw,nosuid,nodev,noexec,mode=0700,size=128m

# Shared base for the three Redis Cluster shards + one-shot init sidecar.
# The anchor absorbs image + network so each service body collapses to its
# hostname + command + (for the seed) volume + healthcheck.
x-redis-node: &redis-node
  image: redis:7
  networks:
    - app-network

services:
  migrate:
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: migrate
    command:
      [
        "sh",
        "-c",
        "prisma generate && python3 scripts/gen_prisma_types_stub.py && prisma migrate deploy",
      ]
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/migrations
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
    <<: *backend-env-files
    environment:
      <<: *backend-env
    networks:
      - app-network
    restart: on-failure
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "prisma migrate status | grep -q 'No pending migrations' || exit 1",
        ]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 5s

  # Three-container 3-master Redis Cluster for local dev. Stock ``redis:7``
  # on each shard runs its default entrypoint; the one-shot ``redis-init``
  # sidecar runs ``redis-cli --cluster create`` once and exits. No volumes
  # — local dev treats the cluster as cache-only, so every ``docker compose
  # up`` starts fresh. (Persisting only the seed volume across restarts is
  # a trap: the other shards come back with new IDs and the seed's
  # ``nodes.conf`` pins stale peers that cluster gossip can't heal.) Each
  # shard announces its own compose hostname via
  # ``--cluster-announce-hostname`` so both in-compose clients (Docker DNS)
  # and host-native clients (handled by the backend's ``address_remap``)
  # receive a consistent address on CLUSTER SLOTS. Ports 17000/17001/17002
  # sit well clear of FalkorDB on host 6380.
  redis-0:
    <<: *redis-node
    hostname: redis-0
    command: redis-server --port 17000 --cluster-enabled yes --cluster-config-file nodes.conf --cluster-node-timeout 5000 --cluster-require-full-coverage no --cluster-announce-hostname redis-0 --cluster-announce-port 17000 --cluster-announce-bus-port 27000 --cluster-preferred-endpoint-type hostname
    ports:
      - "17000:17000"
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "redis-cli -p 17000 ping && redis-cli -p 17000 cluster info | grep -q 'cluster_state:ok'",
        ]
      interval: 10s
      timeout: 5s
      retries: 30
  redis-1:
    <<: *redis-node
    hostname: redis-1
    command: redis-server --port 17001 --cluster-enabled yes --cluster-config-file nodes.conf --cluster-node-timeout 5000 --cluster-require-full-coverage no --cluster-announce-hostname redis-1 --cluster-announce-port 17001 --cluster-announce-bus-port 27001 --cluster-preferred-endpoint-type hostname
    ports:
      - "17001:17001"
  redis-2:
    <<: *redis-node
    hostname: redis-2
    command: redis-server --port 17002 --cluster-enabled yes --cluster-config-file nodes.conf --cluster-node-timeout 5000 --cluster-require-full-coverage no --cluster-announce-hostname redis-2 --cluster-announce-port 17002 --cluster-announce-bus-port 27002 --cluster-preferred-endpoint-type hostname
    ports:
      - "17002:17002"
  # The ``|| cluster create`` short-circuit makes this idempotent within the
  # lifetime of a single compose project: a re-``up`` without ``down`` leaves
  # the already-formed cluster untouched. A ``down`` + ``up`` wipes every
  # shard (no volumes) and re-creates from scratch.
  redis-init:
    <<: *redis-node
    depends_on: [redis-0, redis-1, redis-2]
    restart: "no"
    # A started shard can take a second to accept connections, and this runs once: wait for every PONG.
    command: >-
      sh -c "for node in redis-0:17000 redis-1:17001 redis-2:17002; do
      i=0; until redis-cli -h $${node%:*} -p $${node#*:} ping 2>/dev/null | grep -q PONG || [ $$i -ge 60 ]; do i=$$((i + 1)); sleep 0.5; done; done;
      redis-cli -h redis-0 -p 17000 cluster info 2>/dev/null | grep -q cluster_state:ok ||
      redis-cli --cluster create redis-0:17000 redis-1:17001 redis-2:17002 --cluster-replicas 0 --cluster-yes"
  falkordb:
    image: falkordb/falkordb:latest
    ports:
      - "6380:6379" # FalkorDB Redis protocol (6380 to avoid clash with Redis on 6379)
      - "3001:3000" # FalkorDB web UI
    <<: *backend-env-files
    # `:?` exits the container if the var is unset/empty, surfacing a
    # missing-config error instead of silent `--requirepass ""` auth.
    entrypoint:
      - /bin/sh
      - -c
      - >-
        export REDIS_ARGS="--requirepass $${GRAPHITI_FALKORDB_PASSWORD:?GRAPHITI_FALKORDB_PASSWORD must be set}"
        && exec /var/lib/falkordb/bin/run.sh "$$@"
      - --
    volumes:
      - falkordb_data:/data
    networks:
      - app-network
    healthcheck:
      test:
        [
          "CMD-SHELL",
          'redis-cli -p 6379 -a "$${GRAPHITI_FALKORDB_PASSWORD:-}" --no-auth-warning ping | grep -q PONG',
        ]
      interval: 10s
      timeout: 5s
      retries: 5

  rabbitmq:
    image: rabbitmq:4.1.4
    container_name: rabbitmq
    healthcheck:
      test: rabbitmq-diagnostics -q ping
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 10s
    <<: *backend-env-files
    environment:
      <<: *backend-env
    ports:
      - "5672:5672"

  # One-shot: publishes the public skills catalog (marketplace skills and the
  # expert roster) once the schema is migrated, the same way a deploy does
  # after `prisma migrate deploy`. Idempotent, so a restart that changes
  # nothing is a no-op. Needs GitHub; offline, set SKILLS_CATALOG_PATH in
  # backend/.env to a local checkout. Fails once and visibly rather than
  # looping, and nothing else waits on it: copies catch up lazily.
  publish_skills:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["publish-skills-catalog", "--skip-missing-preloads"]
    depends_on:
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      redis-0:
        condition: service_healthy
    <<: *backend-env-files
    environment:
      <<: *backend-env
    networks:
      - app-network
    restart: "no"

  rest_server:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["rest"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      redis-0:
        condition: service_healthy
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      rabbitmq:
        condition: service_healthy
    <<: *backend-env-files
    environment:
      <<: *backend-env
      CODEX_TEMP_ROOT: /run/autogpt-codex
    tmpfs: *codex-tmpfs
    ports:
      - "8006:8006"
    volumes:
      - workspace-data:/app/autogpt_platform/backend/workspaces
      - store-media-data:/app/autogpt_platform/backend/store-media
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  executor:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["executor"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      redis-0:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
      CODEX_TEMP_ROOT: /run/autogpt-codex
    tmpfs: *codex-tmpfs
    # Metrics only. Reachable from the other containers, not from the host.
    expose:
      - "8002"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  copilot_executor:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["python", "-u", "-m", "backend.copilot.executor"]
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      redis-0:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
      CODEX_TEMP_ROOT: /run/autogpt-codex
      PYTHONUNBUFFERED: "1"
    tmpfs: *codex-tmpfs
    # Metrics only. Reachable from the other containers, not from the host.
    expose:
      - "8008"
    volumes:
      - workspace-data:/app/autogpt_platform/backend/workspaces
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  websocket_server:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["ws"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
      redis-0:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
    ports:
      - "8001:8001"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  database_manager:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["db"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
    <<: *backend-env-files
    environment:
      <<: *backend-env
    # Internal service. Reachable from the other containers, not from the host.
    expose:
      - "8005"
    volumes:
      - store-media-data:/app/autogpt_platform/backend/store-media
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  scheduler_server:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["scheduler"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
      redis-0:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
    # Internal service. Reachable from the other containers, not from the host.
    expose:
      - "8003"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  notification_server:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["notification"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
    # Internal service. Reachable from the other containers, not from the host.
    expose:
      - "8007"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
  platform_linking_manager:
    image: ${COMPOSE_PROJECT_NAME:-autogpt_platform}-backend:latest
    profiles: ["bot"]
    build:
      context: ../
      dockerfile: autogpt_platform/backend/Dockerfile
      target: server
    command: ["platform-linking-manager"] # points to entry in [tool.poetry.scripts] in pyproject.toml
    develop:
      watch:
        - path: ./
          target: autogpt_platform/backend/
          action: rebuild
    depends_on:
      db:
        condition: service_healthy
      redis-0:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
      database_manager:
        condition: service_started
    <<: *backend-env-files
    environment:
      <<: *backend-env
    # Internal service. Reachable from the other containers, not from the host.
    expose:
      - "8009"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
  frontend:
    build:
      context: ../
      dockerfile: autogpt_platform/frontend/Dockerfile
      target: prod
      args:
        NEXT_PUBLIC_PW_TEST: ${NEXT_PUBLIC_PW_TEST:-false}
        NEXT_PUBLIC_FRONTEND_BASE_URL: ${NEXT_PUBLIC_FRONTEND_BASE_URL:-http://localhost:3000}
    depends_on:
      db:
        condition: service_healthy
      migrate:
        condition: service_completed_successfully
    ports:
      - "3000:3000"
    networks:
      - app-network
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    # Load environment variables in order (later overrides earlier)
    env_file:
      - path: ./frontend/.env.default # Base defaults (always exists)
      - path: ./frontend/.env # User overrides (optional)
        required: false
    environment:
      # Server-side environment variables (Docker service names)
      # These override the localhost URLs from env files when running in Docker
      AGPT_SERVER_URL: http://rest_server:8006/api
      AGPT_WS_SERVER_URL: ws://websocket_server:8001/ws
      NEXT_PUBLIC_FRONTEND_BASE_URL: ${NEXT_PUBLIC_FRONTEND_BASE_URL:-http://localhost:3000}
      # Better Auth (embedded in this Next.js server) — talks to Postgres
      # directly; tables live in the platform schema (Prisma migrations).
      DATABASE_URL: postgresql://postgres:your-super-secret-and-long-postgres-password@db:5432/postgres
      AUTH_DB_SCHEMA: platform
      # Node closes an idle keep-alive socket after 5 s (+1 s buffer); a client reusing one right then
      # gets "socket hang up". Keep idle sockets longer than clients and proxies leave them unused.
      KEEP_ALIVE_TIMEOUT: "65000"
volumes:
  workspace-data:
  store-media-data:
  falkordb_data:

networks:
  app-network:
    driver: bridge