docker-compose.dev.yaml

# ============================================================================
# CAIPE DOCKER COMPOSE DEV CONFIGURATION
# ============================================================================
# Development version of docker-compose.yaml with local builds and file mounts.
# For setup instructions and agent orchestration details, refer to:
#   https://cnoe-io.github.io/ai-platform-engineering/getting-started/quick-start
#
# Unified supervisor service controlled by DISTRIBUTED_AGENTS env var:
#
#   All-in-one (all-in-one) — MCP tools in-process, all agents in one container:
#     docker compose -f docker-compose.dev.yaml --profile caipe-supervisor --profile caipe-mongodb up --build
#     Set: DISTRIBUTED_AGENTS= (empty, default)
#
#   Fully distributed — separate A2A agent containers:
#     docker compose -f docker-compose.dev.yaml --profile caipe-supervisor --profile <agent-profiles> up --build
#     Set: DISTRIBUTED_AGENTS=all
#
#   Hybrid — mix of in-process and distributed agents:
#     docker compose -f docker-compose.dev.yaml --profile caipe-supervisor --profile <agent-profiles> up --build
#     Set: DISTRIBUTED_AGENTS=argocd,github  (only listed agents are remote)
#
#   Single-node MCP-only — supervisor runs agents in-process and only external MCP
#   server containers are started:
#     DISTRIBUTED_AGENTS= docker compose -f docker-compose.dev.yaml --profile caipe-supervisor --profile mcp-servers up --build
#
# Agent selection and configuration are controlled through environment variables
# in the .env file—see documentation for full list and descriptions.
# ============================================================================

services:

  ####################################################################################################
  #                              CAIPE Supervisor Deep Agent (Unified)                              #
  ####################################################################################################
  # Unified supervisor: mode is controlled by DISTRIBUTED_AGENTS env var, not separate services.
  #
  #   All-in-one:     DISTRIBUTED_AGENTS= (empty)            — all agents run in-process via MCP
  #   Fully distributed:  DISTRIBUTED_AGENTS=all             — all agents run as remote A2A containers
  #   Hybrid:         DISTRIBUTED_AGENTS=argocd,jira         — only listed agents are remote
  #
  # Start with: docker compose --profile caipe-supervisor [--profile <agent-profiles>] up --build
  caipe-supervisor:
    build:
      context: .
      dockerfile: build/Dockerfile
    image: ghcr.io/cnoe-io/ai-platform-engineering:${IMAGE_TAG:-localtag}
    container_name: caipe-supervisor
    volumes:
      - ${PROMPT_CONFIG_PATH:-./charts/ai-platform-engineering/data/prompt_config.deep_agent.yaml}:/app/prompt_config.yaml
      - ${PROMPT_CONFIG_RAG_PATH:-./charts/ai-platform-engineering/data/prompt_config.rag.yaml}:/app/prompt_config.rag.yaml
      - ${TASK_CONFIG_PATH:-./charts/ai-platform-engineering/data/task_config.yaml}:/app/task_config.yaml
      - ./ai_platform_engineering:/app/ai_platform_engineering
      - ${HOME}/.aws:/home/appuser/.aws:ro
      # Spec 102 — PDP-unavailable fallback. Defaults to /dev/null so the mount
      # is a harmless no-op in dev; the e2e lane sets RBAC_FALLBACK_FILE to the
      # real path on the host (./deploy/keycloak/realm-config-extras.json) AND
      # sets RBAC_FALLBACK_CONFIG_PATH so the helper actually reads it.
      - ${RBAC_FALLBACK_FILE:-/dev/null}:/etc/keycloak/realm-config-extras.json:ro
    env_file:
      - .env
    ports:
      # Host port is parameterized so the e2e lane (spec 102) can publish on a
      # non-conflicting port without a separate compose override file.
      # Default keeps dev behavior unchanged (8000); e2e sets SUPERVISOR_HOST_PORT=28000.
      - "${SUPERVISOR_HOST_PORT:-8000}:8000"
    environment:
      # Dev-only: set DEV_HOT_RELOAD=true to enable uvicorn --reload on the
      # bind-mounted source under /app/ai_platform_engineering. Default off.
      - DEV_HOT_RELOAD=${DEV_HOT_RELOAD:-false}
      - AWS_CONFIG_FILE=/home/appuser/.aws/config
      - AWS_SHARED_CREDENTIALS_FILE=/home/appuser/.aws/credentials
      # Spec 102 — PDP-unavailable fallback. Path is a no-op unless the e2e lane
      # mounts /etc/keycloak/realm-config-extras.json (see volumes below).
      - RBAC_FALLBACK_CONFIG_PATH=${RBAC_FALLBACK_CONFIG_PATH:-}
      # Prompt configuration file path
      - PROMPT_CONFIG_PATH=/app/prompt_config.yaml
      # Task config file path (self-service workflows)
      - TASK_CONFIG_PATH=/app/task_config.yaml

      # MongoDB - primary source for task configs (UI writes here, backend reads)
      - MONGODB_URI=mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@caipe-mongodb:27017/${MONGODB_DATABASE:-caipe}?authSource=admin
      - MONGODB_DATABASE=${MONGODB_DATABASE:-caipe}
      # Disable cache so UI edits are reflected immediately in the agent
      - TASK_CONFIG_CACHE_TTL=${TASK_CONFIG_CACHE_TTL:-0}

      # A2A transport configuration (p2p or slim)
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}

      # Slim dataplane endpoint configuration
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}

      # Agent connectivity configuration
      - AGENT_CONNECTIVITY_ENABLE_BACKGROUND=${AGENT_CONNECTIVITY_ENABLE_BACKGROUND:-true}
      - ENABLE_ARTIFACT_STREAMING=true
      - USE_STRUCTURED_RESPONSE=${USE_STRUCTURED_RESPONSE:-true}
      - AGENT_GATEWAY_URL=${AGENT_GATEWAY_URL:-http://agentgateway:4000}
      # OBO token exchange — supervisor exchanges user JWT for OBO JWT via Keycloak (FR-038d).
      # Use the correctly-namespaced KEYCLOAK_SUPERVISOR_CLIENT_SECRET interpolation (NOT
      # KEYCLOAK_ADMIN_CLIENT_SECRET — that's the UI BFF's var and sharing it would create
      # the same kind of namespace collision that broke slack-bot in 098).
      - KEYCLOAK_SUPERVISOR_CLIENT_ID=${KEYCLOAK_SUPERVISOR_CLIENT_ID:-caipe-platform}
      - KEYCLOAK_SUPERVISOR_CLIENT_SECRET=${KEYCLOAK_SUPERVISOR_CLIENT_SECRET:-caipe-platform-dev-secret}

      # Per-agent distribution control (comma-separated list, or "all" for fully distributed).
      # Leave empty for all-in-one mode (all agents in-process via MCP).
      # Examples: DISTRIBUTED_AGENTS=argocd,aws  or  DISTRIBUTED_AGENTS=all
      - DISTRIBUTED_AGENTS=${DISTRIBUTED_AGENTS:-}

      # Remote A2A agent hostname mappings (used when agents are distributed)
      - ARGOCD_AGENT_HOST=agent-argocd
      - AWS_AGENT_HOST=agent-aws
      - BACKSTAGE_AGENT_HOST=agent-backstage
      - CONFLUENCE_AGENT_HOST=agent-confluence
      - GITLAB_AGENT_HOST=agent-gitlab
      - GITHUB_AGENT_HOST=agent-github
      - JIRA_AGENT_HOST=agent-jira
      - KOMODOR_AGENT_HOST=agent-komodor
      - PAGERDUTY_AGENT_HOST=agent-pagerduty
      - PETSTORE_AGENT_HOST=agent-petstore
      - SLACK_AGENT_HOST=agent-slack
      - SPLUNK_AGENT_HOST=agent-splunk
      - WEATHER_AGENT_HOST=agent-weather
      - WEATHER_AGENT_PORT=8000
      - VICTOROPS_AGENT_HOST=agent-victorops
      - WEBEX_AGENT_HOST=agent-webex
      - NETUTILS_AGENT_HOST=agent-netutils
      - JARVIS_AGENT_HOST=agent-jarvis
      - JARVIS_AGENT_PORT=58000

      # MCP server hostname mappings for in-process (all-in-one) mode.
      # When agents run in-process, they connect to MCP servers by container name.
      - ARGOCD_MCP_HOST=mcp-argocd
      - BACKSTAGE_MCP_HOST=mcp-backstage
      - CONFLUENCE_MCP_HOST=mcp-confluence
      - JIRA_MCP_HOST=mcp-jira
      - KOMODOR_MCP_HOST=mcp-komodor
      - PAGERDUTY_MCP_HOST=mcp-pagerduty
      - SLACK_MCP_HOST=mcp-slack
      - SLACK_MCP_PORT=3001
      - SPLUNK_MCP_HOST=mcp-splunk
      - WEBEX_MCP_HOST=mcp-webex

      # AWS multi-account support
      - DEFAULT_AWS_ACCOUNT_ID=${DEFAULT_AWS_ACCOUNT_ID:-}
      - AWS_ACCOUNT_LIST=${AWS_ACCOUNT_LIST:-}
      - CROSS_ACCOUNT_ROLE_NAME=${CROSS_ACCOUNT_ROLE_NAME:-}

      # The following ENABLE_* environment variables control which agents are active.
      # In .env file, set each variable to 'true' to enable the corresponding agent.
      - ENABLE_ARGOCD=${ENABLE_ARGOCD:-false}
      - ENABLE_AWS=${ENABLE_AWS:-false}
      - ENABLE_BACKSTAGE=${ENABLE_BACKSTAGE:-false}
      - ENABLE_CONFLUENCE=${ENABLE_CONFLUENCE:-false}
      - ENABLE_GITHUB=${ENABLE_GITHUB:-false}
      - ENABLE_GITLAB=${ENABLE_GITLAB:-false}
      - ENABLE_JIRA=${ENABLE_JIRA:-false}
      - ENABLE_KOMODOR=${ENABLE_KOMODOR:-false}
      - ENABLE_PAGERDUTY=${ENABLE_PAGERDUTY:-false}
      - ENABLE_PETSTORE=${ENABLE_PETSTORE:-false}
      - ENABLE_RAG=${ENABLE_RAG:-false}
      - ENABLE_SLACK=${ENABLE_SLACK:-false}
      - ENABLE_SPLUNK=${ENABLE_SPLUNK:-false}
      - ENABLE_VICTOROPS=${ENABLE_VICTOROPS:-false}
      - ENABLE_WEBEX=${ENABLE_WEBEX:-false}
      - ENABLE_WEATHER=${ENABLE_WEATHER:-false}
      - ENABLE_NETUTILS=${ENABLE_NETUTILS:-false}
      - ENABLE_JARVIS=${ENABLE_JARVIS:-false}

      # RAG Service Configuration
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - FETCH_DOCUMENT_MAX_CALLS=${FETCH_DOCUMENT_MAX_CALLS:-5}
      - SEARCH_MAX_CALLS=${SEARCH_MAX_CALLS:-3}
      - RAG_MAX_OUTPUT_CHARS=${RAG_MAX_OUTPUT_CHARS:-10000}
      - RAG_MAX_SEARCH_RESULTS=${RAG_MAX_SEARCH_RESULTS:-3}
      - LANGGRAPH_RECURSION_LIMIT=${LANGGRAPH_RECURSION_LIMIT:-500}

      # GitHub Token for git operations
      - GITHUB_TOKEN=${GITHUB_PERSONAL_ACCESS_TOKEN}

      # Cross-Thread Store Configuration
      - LANGGRAPH_STORE_TYPE=${LANGGRAPH_STORE_TYPE:-memory}
      - LANGGRAPH_STORE_REDIS_URL=${LANGGRAPH_STORE_REDIS_URL:-}
      - LANGGRAPH_STORE_POSTGRES_DSN=${LANGGRAPH_STORE_POSTGRES_DSN:-}
      - LANGGRAPH_STORE_MONGODB_URI=${LANGGRAPH_STORE_MONGODB_URI:-}
      - LANGGRAPH_STORE_TTL_MINUTES=${LANGGRAPH_STORE_TTL_MINUTES:-10080}
      # Embeddings for semantic memory search (shared with RAG stack via EMBEDDINGS_*)
      # LANGGRAPH_STORE_EMBEDDINGS_* overrides take precedence over EMBEDDINGS_*
      - EMBEDDINGS_PROVIDER=${EMBEDDINGS_PROVIDER:-}
      - EMBEDDINGS_MODEL=${EMBEDDINGS_MODEL:-}
      - LANGGRAPH_STORE_EMBEDDINGS_DIMS=${LANGGRAPH_STORE_EMBEDDINGS_DIMS:-}
      - ENABLE_FACT_EXTRACTION=${ENABLE_FACT_EXTRACTION:-false}
      - FACT_EXTRACTION_MODEL=${FACT_EXTRACTION_MODEL:-}

      # In-Thread Checkpoint Persistence Configuration
      - LANGGRAPH_CHECKPOINT_TYPE=${LANGGRAPH_CHECKPOINT_TYPE:-memory}
      - LANGGRAPH_CHECKPOINT_REDIS_URL=${LANGGRAPH_CHECKPOINT_REDIS_URL:-}
      - LANGGRAPH_CHECKPOINT_POSTGRES_DSN=${LANGGRAPH_CHECKPOINT_POSTGRES_DSN:-}
      - LANGGRAPH_CHECKPOINT_MONGODB_URI=${LANGGRAPH_CHECKPOINT_MONGODB_URI:-}
      - LANGGRAPH_CHECKPOINT_MONGODB_DB_NAME=${LANGGRAPH_CHECKPOINT_MONGODB_DB_NAME:-}
      - LANGGRAPH_CHECKPOINT_MONGODB_COLLECTION=${LANGGRAPH_CHECKPOINT_MONGODB_COLLECTION:-}
      - LANGGRAPH_CHECKPOINT_MONGODB_WRITES_COLLECTION=${LANGGRAPH_CHECKPOINT_MONGODB_WRITES_COLLECTION:-}
      - LANGGRAPH_CHECKPOINT_TTL_MINUTES=${LANGGRAPH_CHECKPOINT_TTL_MINUTES:-0}
      # REDIS_URL is required by redisvl (used internally by langgraph-checkpoint-redis)
      - REDIS_URL=${LANGGRAPH_CHECKPOINT_REDIS_URL:-}

      # Self-service task configuration
      - DEFAULT_GITHUB_ORG=${DEFAULT_GITHUB_ORG:-}
      - GITHUB_ORGS=${GITHUB_ORGS:-}
      - GROUPS_AUTOMATION_REPO=${GROUPS_AUTOMATION_REPO:-}
      - GROUPS_TEAMS_PATH=${GROUPS_TEAMS_PATH:-}
      - ORG_MEMBERSHIP_FILE_PATTERN=${ORG_MEMBERSHIP_FILE_PATTERN:-}
      - EMAIL_DOMAIN=${EMAIL_DOMAIN:-}
      - JARVIS_WORKFLOWS_REPO=${JARVIS_WORKFLOWS_REPO:-}
      - ADMIN_REPO=${ADMIN_REPO:-}
      - TERRAFORM_INFRA_REPO=${TERRAFORM_INFRA_REPO:-}
      - JIRA_ASSIGNEE=${JIRA_ASSIGNEE:-}
      - WEBEX_ROOM_ID=${WEBEX_ROOM_ID:-}
      - DEFAULT_CLUSTER_NAME=${DEFAULT_CLUSTER_NAME:-}
      - DEFAULT_CLUSTER_ADDRESS=${DEFAULT_CLUSTER_ADDRESS:-}

      # Per-subagent LLM model overrides (provider:model-name format)
      - OPENAI_API_VERSION=${AZURE_OPENAI_API_VERSION:-2025-03-01-preview}

      # Vault configuration for secure LLM key storage
      - VAULT_ADDR=${VAULT_ADDR:-}
      - VAULT_NAMESPACE=${VAULT_NAMESPACE:-}
      - VAULT_ROLE_ID=${VAULT_ROLE_ID:-}
      - VAULT_SECRET_ID=${VAULT_SECRET_ID:-}
      - VAULT_MOUNT_POINT=${VAULT_MOUNT_POINT:-secret}
      - VAULT_APPROLE_PATH=${VAULT_APPROLE_PATH:-approle}
      - VAULT_PATH_PREFIX=${VAULT_PATH_PREFIX:-}

      # Tracing configuration (will be read from .env if ENABLE_TRACING=true)
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-NOT_SET}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-NOT_SET}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - LANGFUSE_SESSION_ID=${LANGFUSE_SESSION_ID:-ai-platform-engineering}
      - LANGFUSE_USER_ID=${LANGFUSE_USER_ID:-platform-engineer}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
    depends_on:
      caipe-mongodb:
        condition: service_healthy
    command: platform-engineer
    profiles:
      - caipe-supervisor

  ####################################################################################################
  #                                          MongoDB                                                 #
  ####################################################################################################
  # MongoDB for CAIPE UI chat history and user settings
  # Note: Default password "changeme" is for local development only. For production, use
  # the Helm chart and External Secrets to store the MongoDB secret in an external secrets manager.
  caipe-mongodb:
    image: mongo:7.0
    container_name: caipe-mongodb-dev
    ports:
      # Host port is parameterized for the spec 102 e2e lane (sets MONGODB_HOST_PORT=28017)
      # to avoid collision with a host-side MongoDB on 27017. Container port is unchanged.
      - "${MONGODB_HOST_PORT:-27017}:27017"
    environment:
      - MONGO_INITDB_ROOT_USERNAME=${MONGODB_ROOT_USERNAME:-admin}
      - MONGO_INITDB_ROOT_PASSWORD=${MONGODB_ROOT_PASSWORD:-changeme}
      - MONGO_INITDB_DATABASE=${MONGODB_DATABASE:-caipe}
    volumes:
      - mongodb_data:/data/db
      - mongodb_config:/data/configdb
    command: ["mongod", "--quiet", "--logpath", "/dev/null"]
    healthcheck:
      test: ["CMD", "mongosh", "--eval", "db.adminCommand('ping')"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 40s
    profiles:
      - caipe-mongodb
      - caipe-supervisor
      - deps

  # MongoDB Admin UI
  caipe-ui-mongo-express:
    image: mongo-express:latest
    container_name: caipe-mongo-express-dev
    ports:
      - "8081:8081"
    environment:
      - ME_CONFIG_MONGODB_ADMINUSERNAME=${MONGODB_ROOT_USERNAME:-admin}
      - ME_CONFIG_MONGODB_ADMINPASSWORD=${MONGODB_ROOT_PASSWORD:-changeme}
      - ME_CONFIG_MONGODB_URL=mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@mongodb:27017/
      - ME_CONFIG_BASICAUTH=false
    profiles:
      - caipe-ui-mongo-express

  ####################################################################################################
  #                                      SLIM Transport                                              #
  ####################################################################################################
  slim-dataplane:
    image: ghcr.io/agntcy/slim:0.6.1
    container_name: slim-dataplane
    ports: ["46357:46357"]
    environment:
      - PASSWORD=${SLIM_GATEWAY_PASSWORD:-dummy_password}
      - CONFIG_PATH=/config.yaml
    volumes: ["./slim-config.yaml:/config.yaml"]
    command: ["/slim", "--config", "/config.yaml"]
    profiles:
      - slim

  slim-control-plane:
    image: ghcr.io/agntcy/slim/control-plane:0.0.1
    container_name: slim-control-plane
    ports: ["50051:50051", "50052:50052"]
    environment:
      - PASSWORD=${SLIM_GATEWAY_PASSWORD:-dummy_password}
      - CONFIG_PATH=/config.yaml
    volumes: ["./slim-config.yaml:/config.yaml"]
    command: ["/slim", "--config", "/config.yaml"]
    profiles:
      - slim

  ####################################################################################################
  #                                         Sub-Agents                                               #
  ####################################################################################################
  # Shared Dockerfile.a2a / Dockerfile.mcp no longer install aws-cli / kubectl /
  # gh / glab by default to keep agent images small and free of transitive CVEs
  # (e.g. Python bundled inside aws-cli). Per-agent opt-ins live in
  # .github/agents.json under `agent_build_args` for CI; each service below
  # mirrors the same flags in its `args:` block so `docker compose build`
  # produces the same image. Valid flags: INSTALL_AWS_CLI, INSTALL_KUBECTL,
  # INSTALL_GH_CLI, INSTALL_GLAB_CLI — all default to false.
  #
  # AWS Agent
  agent-aws:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=aws
        - INSTALL_AWS_CLI=true
        - INSTALL_KUBECTL=true
    image: ghcr.io/cnoe-io/agent-aws:${IMAGE_TAG:-localtag}
    container_name: agent-aws
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.aws_agent.yaml:/app/prompt_config.aws_agent.yaml
      # Mount only agent_aws subdirectory to preserve .venv from build
      - ./ai_platform_engineering/agents/aws/agent_aws:/app/ai_platform_engineering/agents/agent/agent_aws
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8012:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
      - AWS_AGENT_BACKEND=${AWS_AGENT_BACKEND:-langgraph}
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
      - AWS_REGION=${AWS_REGION}
      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
      - ENABLE_EKS_MCP=${ENABLE_EKS_MCP:-false}
      - ENABLE_COST_EXPLORER_MCP=${ENABLE_COST_EXPLORER_MCP:-false}
      - ENABLE_IAM_MCP=${ENABLE_IAM_MCP:-false}
      - IAM_MCP_READONLY=${IAM_MCP_READONLY:-true}
      - USE_AWS_CLI_AS_TOOL=${USE_AWS_CLI_AS_TOOL:-true}
      # Multi-account support - format: "name1:id1,name2:id2" or just "id1,id2"
      - DEFAULT_AWS_ACCOUNT_ID=${DEFAULT_AWS_ACCOUNT_ID:-common-dev:123456789012}
      - AWS_ACCOUNT_LIST=${AWS_ACCOUNT_LIST:-account1:123456789012,account2:123456789013}
      - CROSS_ACCOUNT_ROLE_NAME=${CROSS_ACCOUNT_ROLE_NAME:-caipe-read-only}
      - STRANDS_LOG_LEVEL=${STRANDS_LOG_LEVEL:-INFO}
      - FASTMCP_LOG_LEVEL=${FASTMCP_LOG_LEVEL:-ERROR}
    profiles:
      - aws
      - all-agents

  # Petstore Agent
  agent-petstore:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=template
        - AGENT_PACKAGE=petstore
    image: ghcr.io/cnoe-io/agent-template:${IMAGE_TAG:-localtag}
    container_name: agent-petstore
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.petstore_agent.yaml:/app/prompt_config.petstore_agent.yaml
      # Mount template agent directory (petstore uses template)
      - ./ai_platform_engineering/agents/template/agent_petstore:/app/ai_platform_engineering/agents/agent/agent_petstore
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8013:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_PORT=443
      - MCP_HOST=petstore.outshift.io
      - PETSTORE_API_KEY=${PETSTORE_API_KEY}
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
    profiles:
      - petstore
      - all-agents

  # GitHub
  # Official GitHub MCP server (HTTP mode) — provides full GitHub tool coverage
  # via the github-mcp-server Go binary served over streamable-http.
  # agent-github connects to this via GITHUB_MCP_HOST/PORT instead of localhost.
  github-mcp-server:
    image: ghcr.io/github/github-mcp-server:latest
    container_name: github-mcp-server
    command: ["http", "--port", "8082", "--toolsets", "all"]
    environment:
      - GITHUB_PERSONAL_ACCESS_TOKEN=${GITHUB_PERSONAL_ACCESS_TOKEN}
    ports: ["8082:8082"]
    profiles:
      - github
      - all-agents
      - mcp-servers

  agent-github:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=github
        - INSTALL_GH_CLI=true
    image: ghcr.io/cnoe-io/agent-github:${IMAGE_TAG:-localtag}
    container_name: agent-github
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.github_agent.yaml:/app/prompt_config.github_agent.yaml
      - ./ai_platform_engineering/agents/github/agent_github:/app/ai_platform_engineering/agents/agent/agent_github
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8001:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - GITHUB_MCP_HOST=github-mcp-server
      - GITHUB_MCP_PORT=8082
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
      - GH_TOKEN=${GITHUB_PERSONAL_ACCESS_TOKEN}  # gh CLI authentication (PAT fallback)
      # GitHub App authentication (recommended - auto-refreshing tokens)
      - GITHUB_APP_ID=${GITHUB_APP_ID:-}
      - GITHUB_APP_PRIVATE_KEY=${GITHUB_APP_PRIVATE_KEY:-}
      - GITHUB_APP_PRIVATE_KEY_PATH=${GITHUB_APP_PRIVATE_KEY_PATH:-}
      - GITHUB_APP_INSTALLATION_ID=${GITHUB_APP_INSTALLATION_ID:-}
    depends_on:
      - github-mcp-server
    profiles:
      - github
      - all-agents

  # GitLab
  agent-gitlab:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=gitlab # key for tracing
        - INSTALL_GLAB_CLI=true
    image: ghcr.io/cnoe-io/agent-gitlab:${IMAGE_TAG:-localtag}
    container_name: agent-gitlab
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.gitlab_agent.yaml:/app/prompt_config.gitlab_agent.yaml
      - ./ai_platform_engineering/agents/gitlab/agent_gitlab:/app/ai_platform_engineering/agents/agent/agent_gitlab
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8014:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=stdio
      - MCP_HOST=mcp-gitlab
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
      - GIT_AUTHOR_NAME=${GIT_AUTHOR_NAME:-AI Agent}
      - GIT_AUTHOR_EMAIL=${GIT_AUTHOR_EMAIL:-ai-agent@cnoe.io}
      - GIT_COMMITTER_NAME=${GIT_COMMITTER_NAME:-AI Agent}
      - GIT_COMMITTER_EMAIL=${GIT_COMMITTER_EMAIL:-ai-agent@cnoe.io}
    depends_on:
      mcp-gitlab:
        condition: service_started
    profiles:
      - gitlab

  # GitLab MCP Server (@zereight/mcp-gitlab)
  # Tool filtering is controlled via GITLAB_DENIED_TOOLS_REGEX using prefix matching (^prefix_)
  #
  # Example regex patterns for different permission levels:
  #
  # READ-ONLY (default) - blocks all write operations:
  #   ^(delete_|remove_|create_|fork_|new_|update_|edit_|merge_|push_|publish_|retry_|cancel_|play_|promote_|upload_|resolve_|bulk_)|^(execute_graphql)$
  #
  # ALLOW CREATE (can create MRs, issues, etc.) - blocks delete/update:
  #   ^(delete_|remove_|update_|edit_|merge_|push_|publish_|retry_|cancel_|play_|promote_|upload_|resolve_|bulk_)|^(execute_graphql)$
  #
  # ALLOW UPDATE (can update MRs, issues, etc.) - blocks delete/create:
  #   ^(delete_|remove_|create_|fork_|new_)|^(execute_graphql)$
  #
  # ALLOW CREATE + UPDATE (full write access except delete):
  #   ^(delete_|remove_)|^(execute_graphql)$
  #
  # To block specific tools, add them as exact matches: |^(tool1|tool2)$
  # Example blocking approve/unapprove: ...|^(execute_graphql|approve_merge_request|unapprove_merge_request)$
  #
  # Note: execute_graphql is always blocked in the above examples as it bypasses prefix-based permission controls
  mcp-gitlab:
    image: zereight050/gitlab-mcp:latest
    container_name: mcp-gitlab
    env_file: [.env]
    ports: ["18010:8000"]
    environment:
      - STREAMABLE_HTTP=true
      - HOST=0.0.0.0
      - PORT=8000
      - MAX_SESSIONS=${GITLAB_MCP_MAX_SESSIONS:-1000}
      - MAX_REQUESTS_PER_MINUTE=${GITLAB_MCP_MAX_REQUESTS_PER_MINUTE:-60}
      - USE_PIPELINE=true
      - GITLAB_API_URL=https://${GITLAB_HOST:-gitlab.com}/api/v4
      - GITLAB_PERSONAL_ACCESS_TOKEN=${GITLAB_PERSONAL_ACCESS_TOKEN:-${GITLAB_TOKEN}}
      - REMOTE_AUTHORIZATION=${GITLAB_MCP_REMOTE_AUTHORIZATION:-true}
 
      # Option 1: Simple read-only mode (blocks all write operations)
      - GITLAB_READ_ONLY_MODE=${GITLAB_READ_ONLY_MODE:-true}
      # Option 2: Fine-grained control via regex (see examples in comments above)
      # - GITLAB_DENIED_TOOLS_REGEX=${GITLAB_DENIED_TOOLS_REGEX:-^(delete_|remove_|create_|fork_|new_|update_|edit_|merge_|push_|publish_|retry_|cancel_|play_|promote_|upload_|resolve_|bulk_)|^(execute_graphql)$}
    profiles:
      - gitlab
      - all-agents
      - mcp-servers

  # Weather Agent
  agent-weather:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=weather
    image: ghcr.io/cnoe-io/agent-weather:${IMAGE_TAG:-localtag}
    container_name: agent-weather
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.weather_agent.yaml:/app/prompt_config.weather_agent.yaml
      - ./ai_platform_engineering/agents/weather/agent_weather:/app/ai_platform_engineering/agents/agent/agent_weather
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8002:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=weather.outshift.io
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    profiles:
      - weather
      - all-agents

  # Backstage Agent
  agent-backstage:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=backstage
    image: ghcr.io/cnoe-io/agent-backstage:${IMAGE_TAG:-localtag}
    container_name: agent-backstage
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.backstage_agent.yaml:/app/prompt_config.backstage_agent.yaml
      - ./ai_platform_engineering/agents/backstage/agent_backstage:/app/ai_platform_engineering/agents/agent/agent_backstage
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8003:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-backstage
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-backstage:
        condition: service_healthy
    profiles:
      - backstage
      - all-agents

  mcp-backstage:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=backstage
    image: ghcr.io/cnoe-io/mcp-backstage:${IMAGE_TAG:-localtag}
    container_name: mcp-backstage
    env_file: [.env]
    ports: ["18001:8000"]
    volumes:
      - ./ai_platform_engineering/agents/backstage/mcp/mcp_backstage:/app/mcp_backstage
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - backstage
      - all-agents
      - mcp-servers

  # ArgoCD Agents
  agent-argocd:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=argocd
    image: ghcr.io/cnoe-io/agent-argocd:${IMAGE_TAG:-localtag}
    container_name: agent-argocd
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.argocd_agent.yaml:/app/prompt_config.argocd_agent.yaml
      - ./ai_platform_engineering/agents/argocd/agent_argocd:/app/ai_platform_engineering/agents/agent/agent_argocd
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8004:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-argocd
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-argocd:
        condition: service_healthy
    profiles:
      - argocd
      - all-agents

  mcp-argocd:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=argocd
    image: ghcr.io/cnoe-io/mcp-argocd:${IMAGE_TAG:-localtag}
    container_name: mcp-argocd
    env_file: [.env]
    ports: ["18002:8000"]
    volumes:
      - ./ai_platform_engineering/agents/argocd/mcp/mcp_argocd:/app/mcp_argocd
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - argocd
      - all-agents
      - mcp-servers

  # Confluence Agents
  agent-confluence:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=confluence
    image: ghcr.io/cnoe-io/agent-confluence:${IMAGE_TAG:-localtag}
    container_name: agent-confluence
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.confluence_agent.yaml:/app/prompt_config.confluence_agent.yaml
      - ./ai_platform_engineering/agents/confluence/agent_confluence:/app/ai_platform_engineering/agents/agent/agent_confluence
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8005:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-confluence
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-confluence:
        condition: service_started
    profiles:
      - confluence
      - all-agents

  mcp-confluence:
    image: ghcr.io/sooperset/mcp-atlassian:latest
    container_name: mcp-confluence
    env_file: [.env]
    ports: ["18003:8000"]
    environment:
      - CONFLUENCE_URL=${CONFLUENCE_URL}
      - CONFLUENCE_USERNAME=${CONFLUENCE_USERNAME}
      - CONFLUENCE_API_TOKEN=${CONFLUENCE_API_TOKEN}
    profiles:
      - confluence
      - all-agents
      - mcp-servers

  # Jira Agent
  agent-jira:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=jira
    image: ghcr.io/cnoe-io/agent-jira:${IMAGE_TAG:-localtag}
    container_name: agent-jira
    env_file: [.env]
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.jira_agent.yaml:/app/prompt_config.jira_agent.yaml
      - ./ai_platform_engineering/agents/jira/agent_jira:/app/ai_platform_engineering/agents/agent/agent_jira
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    ports: ["8006:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-jira
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-jira:
        condition: service_healthy
    profiles:
      - jira
      - all-agents

  mcp-jira:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=jira
    image: ghcr.io/cnoe-io/mcp-jira:${IMAGE_TAG:-localtag}
    container_name: mcp-jira
    env_file: [.env]
    ports: ["18004:8000"]
    volumes:
      - ./ai_platform_engineering/agents/jira/mcp/mcp_jira:/app/mcp_jira
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
      # AgentGateway is the PEP for the normal gateway path. The Jira MCP
      # upstream must not require its own OAuth middleware here, because AGW's
      # MCP router does not forward upstream auth per target today. Set
      # JIRA_MCP_AUTH_MODE=oauth2 only when testing the direct MCP endpoint.
      - MCP_AUTH_MODE=${JIRA_MCP_AUTH_MODE:-none}
      - MCP_TRUSTED_LOCALHOST=${JIRA_MCP_TRUSTED_LOCALHOST:-false}
      - JWKS_URI=${JIRA_MCP_JWKS_URI:-http://keycloak:7080/realms/caipe/protocol/openid-connect/certs}
      - AUDIENCE=${JIRA_MCP_AUDIENCE:-caipe-platform}
      - ISSUER=${JIRA_MCP_ISSUER:-http://localhost:7080/realms/caipe}
      # Optional PDP check for embedded / direct-local MCP use. Off by default
      # for the normal AgentGateway path so CEL remains the RBAC source of truth.
      - MCP_PDP_ENABLED=${JIRA_MCP_PDP_ENABLED:-false}
      - MCP_PDP_RESOURCE=${JIRA_MCP_PDP_RESOURCE:-mcp_jira}
      - MCP_PDP_SCOPE=${JIRA_MCP_PDP_SCOPE:-invoke}
      - MCP_PDP_AUDIENCE=${JIRA_MCP_PDP_AUDIENCE:-caipe-platform}
      - MCP_PDP_TOKEN_ENDPOINT=${JIRA_MCP_PDP_TOKEN_ENDPOINT:-http://keycloak:7080/realms/caipe/protocol/openid-connect/token}
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - jira
      - all-agents
      - mcp-servers

  # Komodor Agent
  agent-komodor:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=komodor
    image: ghcr.io/cnoe-io/agent-komodor:${IMAGE_TAG:-localtag}
    container_name: agent-komodor
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.komodor_agent.yaml:/app/prompt_config.komodor_agent.yaml
      - ./ai_platform_engineering/agents/komodor/agent_komodor:/app/ai_platform_engineering/agents/agent/agent_komodor
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8007:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-komodor
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-komodor:
        condition: service_healthy
    profiles:
      - komodor
  mcp-komodor:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=komodor
    image: ghcr.io/cnoe-io/mcp-komodor:${IMAGE_TAG:-localtag}
    container_name: mcp-komodor
    env_file: [.env]
    ports: ["18005:8000"]
    volumes:
      - ./ai_platform_engineering/agents/komodor/mcp/mcp_komodor:/app/mcp_komodor
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - komodor
      - mcp-servers

  # PagerDuty Agent
  agent-pagerduty:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=pagerduty
    image: ghcr.io/cnoe-io/agent-pagerduty:${IMAGE_TAG:-localtag}
    container_name: agent-pagerduty
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.pagerduty_agent.yaml:/app/prompt_config.pagerduty_agent.yaml
      - ./ai_platform_engineering/agents/pagerduty/agent_pagerduty:/app/ai_platform_engineering/agents/agent/agent_pagerduty
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8008:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-pagerduty
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-pagerduty:
        condition: service_healthy
    profiles:
      - pagerduty
      - all-agents

  mcp-pagerduty:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=pagerduty
    image: ghcr.io/cnoe-io/mcp-pagerduty:${IMAGE_TAG:-localtag}
    container_name: mcp-pagerduty
    env_file: [.env]
    ports: ["18006:8000"]
    volumes:
      - ./ai_platform_engineering/agents/pagerduty/mcp/mcp_pagerduty:/app/mcp_pagerduty
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - pagerduty
      - all-agents
      - mcp-servers

  # Slack Agent
  agent-slack:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=slack
    image: ghcr.io/cnoe-io/agent-slack:${IMAGE_TAG:-localtag}
    container_name: agent-slack
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.slack_agent.yaml:/app/prompt_config.slack_agent.yaml
      - ./ai_platform_engineering/agents/slack/agent_slack:/app/ai_platform_engineering/agents/agent/agent_slack
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8009:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-slack
      - MCP_PORT=3001
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-slack:
        condition: service_healthy
    profiles:
      - slack
      - all-agents

  # OSS Slack MCP Server (https://github.com/korotovsky/slack-mcp-server)
  mcp-slack:
    image: ghcr.io/korotovsky/slack-mcp-server:v1.2.3
    container_name: mcp-slack
    env_file: [.env]
    ports: ["18007:3001"]
    environment:
      - SLACK_MCP_HOST=0.0.0.0
      - SLACK_MCP_PORT=3001
      - SLACK_MCP_XOXB_TOKEN=${SLACK_BOT_TOKEN}
      # Message posting (disabled by default for safety). Set to 'true' for all
      # channels, or restrict to specific channel IDs (comma-separated), or use
      # '!' prefix to exclude channels.
      # Examples: 'true' (all), 'C123,C456' (whitelist), '!C789' (blacklist)
      # - SLACK_MCP_ADD_MESSAGE_TOOL=${SLACK_MCP_ADD_MESSAGE_TOOL:-false}
    command: ["--transport", "http"]
    healthcheck:
      test: ["CMD-SHELL", "curl -so /dev/null -w '%{http_code}' --max-time 2 -X POST http://localhost:3001/mcp | grep -q 400"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 60s
    profiles:
      - slack
      - all-agents
      - mcp-servers

  # Splunk
  agent-splunk:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=splunk
    image: ghcr.io/cnoe-io/agent-splunk:${IMAGE_TAG:-localtag}
    container_name: agent-splunk
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.splunk_agent.yaml:/app/prompt_config.splunk_agent.yaml
      - ./ai_platform_engineering/agents/splunk/agent_splunk:/app/ai_platform_engineering/agents/agent/agent_splunk
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8010:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-splunk
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-splunk:
        condition: service_healthy
    profiles:
      - splunk
      - all-agents

  mcp-splunk:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=splunk
    image: ghcr.io/cnoe-io/mcp-splunk:${IMAGE_TAG:-localtag}
    container_name: mcp-splunk
    env_file: [.env]
    ports: ["18008:8000"]
    volumes:
      - ./ai_platform_engineering/agents/splunk/mcp/mcp_splunk:/app/mcp_splunk
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - splunk
      - all-agents
      - mcp-servers

  # Webex
  agent-webex:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=webex
    image: ghcr.io/cnoe-io/agent-webex:${IMAGE_TAG:-localtag}
    container_name: agent-webex
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.webex_agent.yaml:/app/prompt_config.webex_agent.yaml
      - ./ai_platform_engineering/agents/webex/agent_webex:/app/ai_platform_engineering/agents/agent/agent_webex
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8011:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-webex
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-webex:
        condition: service_healthy
    profiles:
      - webex
      - all-agents

  mcp-webex:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=webex
    image: ghcr.io/cnoe-io/mcp-webex:${IMAGE_TAG:-localtag}
    container_name: mcp-webex
    env_file: [.env]
    ports: ["18009:8000"]
    volumes:
      - ./ai_platform_engineering/agents/webex/mcp/mcp_webex:/app/mcp_webex
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - webex
      - all-agents
      - mcp-servers

  # VictorOps Agent
  agent-victorops:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=victorops
    image: ghcr.io/cnoe-io/agent-victorops:${IMAGE_TAG:-localtag}
    container_name: agent-victorops
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.victorops_agent.yaml:/app/prompt_config.victorops_agent.yaml
      - ./ai_platform_engineering/agents/victorops/agent_victorops:/app/ai_platform_engineering/agents/agent/agent_victorops
      - ./ai_platform_engineering/agents/victorops/mcp/mcp_victorops:/app/ai_platform_engineering/agents/agent/mcp/mcp_victorops
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8016:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-victorops
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-victorops:
        condition: service_healthy
    profiles:
      - victorops

  mcp-victorops:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=victorops
    image: ghcr.io/cnoe-io/mcp-victorops:${IMAGE_TAG:-localtag}
    container_name: mcp-victorops
    env_file: [.env]
    ports: ["18012:8000"]
    volumes:
      - ./ai_platform_engineering/agents/victorops/mcp/mcp_victorops:/app/mcp_victorops
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - victorops
      - mcp-servers

  ####################################################################################################
  #                                      NETUTILS AGENT (NetUtils)                                    #
  ####################################################################################################
  # NetUtils agent: DNS, DHCP, dnsmasq management, ping, traceroute, port checks
  # Uses a custom MCP Dockerfile with network diagnostic tools pre-installed
  # Optionally connects to a dnsmasq sidecar for DHCP/DNS server management
  agent-netutils:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.a2a
      args:
        - AGENT_NAME=netutils
    image: ghcr.io/cnoe-io/agent-netutils:${IMAGE_TAG:-localtag}
    container_name: agent-netutils
    volumes:
      - ./charts/ai-platform-engineering/data/prompt_config.netutils_agent.yaml:/app/prompt_config.netutils_agent.yaml
      - ./ai_platform_engineering/agents/netutils/agent_netutils:/app/ai_platform_engineering/agents/agent/agent_netutils
      - ./ai_platform_engineering/utils:/app/ai_platform_engineering/utils
    env_file: [.env]
    ports: ["8015:8000"]
    environment:
      - A2A_TRANSPORT=${A2A_TRANSPORT:-p2p}
      - MCP_MODE=http
      - MCP_HOST=mcp-netutils
      - MCP_PORT=8000
      - SLIM_ENDPOINT=${SLIM_ENDPOINT:-http://slim-dataplane:46357}
    depends_on:
      mcp-netutils:
        condition: service_healthy
    profiles:
      - netutils-agent

  mcp-netutils:
    build:
      context: .
      dockerfile: build/agents/Dockerfile.mcp
      args:
        - AGENT_NAME=netutils
        - INSTALL_NETUTILS=true
    image: ghcr.io/cnoe-io/mcp-netutils:${IMAGE_TAG:-localtag}
    container_name: mcp-netutils
    env_file: [.env]
    ports: ["18011:8000"]
    volumes:
      - ./ai_platform_engineering/agents/netutils/mcp/mcp_netutils:/app/mcp_netutils
    environment:
      - MCP_MODE=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=8000
    healthcheck:
      test: ["CMD", "python", "-c", "import socket; s=socket.socket(); s.settimeout(1); result=s.connect_ex(('localhost', 8000)); s.close(); exit(0 if result == 0 else 1)"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 15s
    profiles:
      - netutils-agent
      - mcp-servers

  ####################################################################################################
  #                                      AGENT JARVIS A2A P2P                                        #
  ####################################################################################################
  # Jarvis is an external A2A agent (pre-built image). It exposes two ports:
  #   - 8000: FastAPI health/agent-card endpoint
  #   - 58000: A2A protocol endpoint (used by the supervisor via JARVIS_AGENT_PORT)
  #
  # Configuration lives in .env.jarvis (NOT committed to git — contains credentials).
  # See docker-compose.jarvis.yaml for the canonical reference.
  agent-jarvis:
    image: jarvis-agent:dev
    container_name: agent-jarvis
    restart: unless-stopped
    env_file:
      - .env.jarvis
    ports:
      - "48000:8000"   # FastAPI port (health, agent-card)
      - "58000:58000"  # A2A protocol port
    environment:
      - A2A_TRANSPORT=p2p
      - ALLOWED_ORIGINS=*
      - LANGGRAPH_CHECKPOINT_MEMORY_SAVER=memory
      - ENABLE_KNOWLEDGE_GRAPH=false
      - ENABLE_K8S_AGENT=false
      - AIGATEWAY_SERVER_URL=${AIGATEWAY_SERVER_URL:-NOT_SET}
      # Tracing (inherits from .env if ENABLE_TRACING=true)
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-NOT_SET}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-NOT_SET}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
    command: python3 -m jarvis_agent.combined_server --fastapi-port 8000 --a2a-port 58000
    profiles:
      - jarvis

  ####################################################################################################
  #                                      CAIPE UI                                                    #
  ####################################################################################################
  # CAIPE UI - A2A Message Visualizer with 3-panel layout
  # Features: Chat history, Chat UI with final output, A2A message stream visualization
  # Includes Use Cases gallery inspired by ag-ui composer
  #
  # Build: Uses build/Dockerfile.caipe-ui (Next.js standalone with Node.js server)
  #
  # Configuration:
  #   CAIPE_URL: URL of the CAIPE supervisor A2A endpoint
  #   - Default (dev): http://localhost:8000
  #   - Default (docker): http://caipe-supervisor:8000
  #   - Override with: CAIPE_URL=http://your-host:port
  # ---------------------------------------------------------------------------
  # caipe-ui (hot/dev) and caipe-ui-prod (prod-parity) are TWO first-class
  # services that share most env/volumes/depends_on but differ in:
  #   - build target (`dev` vs `runner`)
  #   - NODE_ENV (`development` vs `production`)
  #   - command (`npm run dev` vs `/app/entrypoint.sh`)
  #   - image suffix (`-dev` vs `-prod`) so both tags persist on disk
  #   - source bind-mounts (only caipe-ui has ./ui/src etc. for hot reload)
  #   - profile (`caipe-ui` vs `caipe-ui-prod`)
  #
  # Two ways to run them:
  #   make caipe-ui-hot   → boots `caipe-ui`      via profile `caipe-ui`
  #                          (next dev, hot reload, bind-mounted ./ui/src)
  #   make caipe-ui-prod  → boots `caipe-ui-prod` via profile `caipe-ui-prod`
  #                          (next build + next start, NO hot reload)
  #
  # IMPORTANT: both services bind host port 3000 — they are MUTUALLY EXCLUSIVE
  # at runtime. Keycloak's caipe-ui client only allow-lists
  # http://localhost:3000/* as a redirect URI (see deploy/keycloak/realm-config.json),
  # so prod-parity must reuse 3000 too. The Makefile targets bring the sibling
  # down before bringing the chosen one up.
  # ---------------------------------------------------------------------------
  caipe-ui:
    build:
      context: .
      dockerfile: build/Dockerfile.caipe-ui
      target: dev
      args:
        - IMAGE_TAG=${IMAGE_TAG:-localtag}
        - GIT_COMMIT=${GIT_COMMIT:-unknown}
        - BUILD_DATE=${BUILD_DATE:-}
        # Local prod-parity rebuilds skip Next's duplicate typecheck; run
        # `npm run lint` / targeted tests separately when validating changes.
        - CAIPE_UI_FAST_BUILD=${CAIPE_UI_FAST_BUILD:-true}
    image: ghcr.io/cnoe-io/caipe-ui:${IMAGE_TAG:-localtag}-dev
    container_name: caipe-ui
    ports: ["3000:3000"]
    env_file:
      - .env
    environment:
      - NODE_ENV=development
      # Spec 102 — Playwright sets E2E_RUN=true to enable test-only fixtures
      # (see ui/src/lib/test-fixtures.ts). Defaults to false in dev.
      - E2E_RUN=${E2E_RUN:-false}
      # CAIPE supervisor A2A endpoint URL
      # Override with A2A_BASE_URL environment variable
      - A2A_BASE_URL=${A2A_BASE_URL:-http://caipe-supervisor:8000}
      # Enable/Disable RAG Knowledge Bases (default: true)
      - RAG_ENABLED=${RAG_ENABLED:-true}
      # RAG Server URL for knowledge base API (internal Docker network)
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - RAG_URL=${RAG_URL:-http://localhost:9446}
      # NextAuth configuration (required for production)
      # Generate a secret with: openssl rand -base64 32
      - NEXTAUTH_SECRET=${NEXTAUTH_SECRET:-caipe-dev-secret-change-in-production}
      - NEXTAUTH_URL=${NEXTAUTH_URL:-http://localhost:3000}
      # SSO Configuration - reads from .env file
      # The app reads SSO_ENABLED, OIDC_CLIENT_ID, OIDC_CLIENT_SECRET, OIDC_ISSUER directly
      - SSO_ENABLED=${SSO_ENABLED:-false}
      - OIDC_CLIENT_ID=${OIDC_CLIENT_ID:-}
      - OIDC_CLIENT_SECRET=${OIDC_CLIENT_SECRET:-}
      - OIDC_ISSUER=${OIDC_ISSUER:-}
      # Spec 104 — Slack-bot OBO tokens are minted with `aud=agentgateway` so they
      # can hit AGW directly. The same token also flows through the BFF on its
      # way to AGW, so the BFF must accept that audience too. Adding it here
      # via OIDC_EXTRA_AUDIENCES (defaults to "agentgateway") keeps the OBO
      # path working without forcing every operator to set the env var.
      # The RBAC e2e persona fixture mints tokens from the caipe-platform
      # resource-server client, so the dev BFF accepts that audience too.
      # For Okta-style deployments, also see OIDC_ACCEPTED_AUDIENCES.
      - OIDC_EXTRA_AUDIENCES=${OIDC_EXTRA_AUDIENCES:-agentgateway}
      - OIDC_ACCEPTED_AUDIENCES=${OIDC_ACCEPTED_AUDIENCES:-caipe-platform}
      # OpenFGA ReBAC reconciliation is opt-in so non-RBAC local UI runs do
      # not require the OpenFGA profile. The remote RBAC stack enables it in
      # `.env` and writes Team Resources changes as OpenFGA tuples.
      - OPENFGA_RECONCILE_ENABLED=${OPENFGA_RECONCILE_ENABLED:-false}
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      # Keep local .env role defaults from affecting RAG authorization paths.
      - ALLOW_TRUSTED_NETWORK=false
      - TRUSTED_NETWORK_CIDRS=
      - TRUSTED_NETWORK_DEFAULT_ROLE=
      - RBAC_ADMIN_GROUPS=
      - RBAC_DEFAULT_AUTHENTICATED_ROLE=
      - RBAC_READONLY_GROUPS=
      - RBAC_INGESTONLY_GROUPS=
      - AUTHZ_TRACING_ENABLED=${AUTHZ_TRACING_ENABLED:-false}
      - OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-}
      # OIDC_DISCOVERY_URL: Docker-internal URL used for server-side discovery +
      # token refresh. Defaults to the in-network Keycloak URL so caipe-ui can
      # fetch the well-known doc without going via the host. Browser still uses
      # OIDC_ISSUER (localhost:7080) for the auth redirect.
      - OIDC_DISCOVERY_URL=${OIDC_DISCOVERY_URL:-http://keycloak:7080/realms/caipe}
      - OIDC_GROUP_CLAIM=${OIDC_GROUP_CLAIM:-}
      - OIDC_REQUIRED_GROUP=${OIDC_REQUIRED_GROUP:-}
      - OIDC_REQUIRED_ADMIN_GROUP=${OIDC_REQUIRED_ADMIN_GROUP:-}
      - BOOTSTRAP_ADMIN_EMAILS=${BOOTSTRAP_ADMIN_EMAILS:-}
      - IDENTITY_SYNC_LOGIN_CLAIMS_ENABLED=${IDENTITY_SYNC_LOGIN_CLAIMS_ENABLED:-true}
      # Login-time team auto-creation. STRICT opt-in; defaults to "false" to
      # preserve the locked-down login posture. Even with this set to "true",
      # the matched identity-group-sync rule must also have auto_create_team=true
      # for a team to actually be created (defense in depth — see
      # ui/src/lib/rbac/identity-group-sync-planner.ts).
      - IDENTITY_SYNC_LOGIN_AUTO_CREATE_TEAMS=${IDENTITY_SYNC_LOGIN_AUTO_CREATE_TEAMS:-false}
      - CAIPE_ORG_KEY=${CAIPE_ORG_KEY:-caipe}
      - CAIPE_ORG_DISPLAY_NAME=${CAIPE_ORG_DISPLAY_NAME:-CAIPE}
      - OIDC_ENABLE_REFRESH_TOKEN=${OIDC_ENABLE_REFRESH_TOKEN:-true}
      # UI Branding (optional — set in .env or leave blank for CAIPE defaults)
      # No NEXT_PUBLIC_ prefix needed; the server reads process.env at request time
      - APP_NAME=${APP_NAME:-}
      - TAGLINE=${TAGLINE:-}
      - DESCRIPTION=${DESCRIPTION:-}
      - LOGO_URL=${LOGO_URL:-}
      - LOGO_STYLE=${LOGO_STYLE:-}
      - SPINNER_COLOR=${SPINNER_COLOR:-}
      - SHOW_POWERED_BY=${SHOW_POWERED_BY:-}
      - SUPPORT_EMAIL=${SUPPORT_EMAIL:-}
      - PREVIEW_MODE=${PREVIEW_MODE:-}
      # Workflow runner - set WORKFLOW_RUNNER_ENABLED=true to show Run Workflow button and Multi-Step Workflows
      - WORKFLOW_RUNNER_ENABLED=${WORKFLOW_RUNNER_ENABLED:-false}
      # Workflows tab - set WORKFLOWS_ENABLED=true to show the top-level Workflows UI
      - WORKFLOWS_ENABLED=${WORKFLOWS_ENABLED:-false}
      # Dynamic Agents - set DYNAMIC_AGENTS_ENABLED=true and add dynamic-agents profile to enable Custom Agents tab
      - DYNAMIC_AGENTS_ENABLED=${DYNAMIC_AGENTS_ENABLED:-false}
      - DYNAMIC_AGENTS_URL=${DYNAMIC_AGENTS_URL:-http://dynamic-agents:8001}
      # Skill Scanner — when the `skill-scanner` profile is enabled, the UI
      # uploads ZIP'd skill packages to this service for security scanning.
      # If unreachable the UI gracefully reports "scanner unavailable" instead
      # of failing the skill save.
      - SKILL_SCANNER_URL=${SKILL_SCANNER_URL:-http://skill-scanner:8000}
      # Skill Hub GitHub crawl/import token. Keep the value in .env or the shell;
      # never hardcode tokens in compose.
      - GITHUB_TOKEN=${GITHUB_TOKEN:-${GITHUB_PERSONAL_ACCESS_TOKEN:-}}
      # MongoDB (when running with profile caipe-ui; mongodb service is in same profile)
      # (changeme default is for local dev only; for production use Helm chart and External Secrets)
      - MONGODB_URI=${MONGODB_URI:-mongodb://admin:changeme@caipe-mongodb:27017/caipe?authSource=admin}
      - MONGODB_DATABASE=${MONGODB_DATABASE:-caipe}
      # Connections & Secrets. Local OAuth connector bootstrap reads these
      # values from .env; Kubernetes deployments use ESO to inject the same
      # variable names into caipe-ui.
      - CAIPE_CREDENTIALS_ENABLED=${CAIPE_CREDENTIALS_ENABLED:-false}
      - CREDENTIAL_STORE_BACKEND=${CREDENTIAL_STORE_BACKEND:-mongodb-envelope}
      - CREDENTIAL_KEY_PROVIDER=${CREDENTIAL_KEY_PROVIDER:-local-cmk}
      - CREDENTIAL_KMS_CMK_ID=${CREDENTIAL_KMS_CMK_ID:-}
      - CREDENTIAL_KMS_REGION=${CREDENTIAL_KMS_REGION:-}
      # DEV-ONLY escape hatch: lets the local (local-cmk/dev-local) key wrapper
      # run under NODE_ENV=production for prod-parity testing. INSECURE — keys
      # are wrapped with locally-derived material, not a real KMS/HSM. Never set
      # this true in a real production deployment.
      - CREDENTIAL_ALLOW_INSECURE_LOCAL_KEY_WRAP=${CREDENTIAL_ALLOW_INSECURE_LOCAL_KEY_WRAP:-false}
      - CREDENTIAL_SERVICE_AUDIENCE=${CREDENTIAL_SERVICE_AUDIENCE:-caipe-credential-service}
      - CREDENTIAL_BOOTSTRAP_OAUTH_CONNECTORS=${CREDENTIAL_BOOTSTRAP_OAUTH_CONNECTORS:-false}
      - GITHUB_CLIENT_ID=${GITHUB_CLIENT_ID:-}
      - GITHUB_CLIENT_SECRET=${GITHUB_CLIENT_SECRET:-}
      - GITHUB_REDIRECT_URI=${GITHUB_REDIRECT_URI:-}
      - CONFLUENCE_CLIENT_ID=${CONFLUENCE_CLIENT_ID:-}
      - CONFLUENCE_CLIENT_SECRET=${CONFLUENCE_CLIENT_SECRET:-}
      - CONFLUENCE_REDIRECT_URI=${CONFLUENCE_REDIRECT_URI:-}
      - WEBEX_CLIENT_ID=${WEBEX_CLIENT_ID:-}
      - WEBEX_CLIENT_SECRET=${WEBEX_CLIENT_SECRET:-}
      - WEBEX_REDIRECT_URI=${WEBEX_REDIRECT_URI:-}
      - PAGERDUTY_CLIENT_ID=${PAGERDUTY_CLIENT_ID:-}
      - PAGERDUTY_CLIENT_SECRET=${PAGERDUTY_CLIENT_SECRET:-}
      - PAGERDUTY_REDIRECT_URI=${PAGERDUTY_REDIRECT_URI:-}
      - PAGERDUTY_SCOPES=${PAGERDUTY_SCOPES:-}
      # 098 Enterprise RBAC — Keycloak AuthZ (requires --profile rbac)
      - KEYCLOAK_URL=${KEYCLOAK_URL:-http://keycloak:7080}
      - KEYCLOAK_REALM=${KEYCLOAK_REALM:-caipe}
      - KEYCLOAK_RESOURCE_SERVER_ID=${KEYCLOAK_RESOURCE_SERVER_ID:-caipe-platform}
      - KEYCLOAK_CLIENT_SECRET=${KEYCLOAK_CLIENT_SECRET:-caipe-platform-dev-secret}
      - RBAC_CACHE_TTL_SECONDS=${RBAC_CACHE_TTL_SECONDS:-60}
      # 098 Enterprise RBAC — Keycloak Admin API (US6: role/mapping CRUD from Admin UI)
      - KEYCLOAK_ADMIN_CLIENT_ID=${KEYCLOAK_ADMIN_CLIENT_ID:-admin-cli}
      - KEYCLOAK_ADMIN_CLIENT_SECRET=${KEYCLOAK_ADMIN_CLIENT_SECRET:-}
      # Production-safety gate (R1) for the BFF admin/admin fallback in
      # `ui/src/lib/rbac/keycloak-admin.ts`. Default-on in dev so a fresh
      # docker-compose stack still boots before the operator wires a real
      # service-account secret. NEVER set this to "true" in a production
      # chart values file — see the comment in
      # charts/.../caipe-ui/values.yaml.
      - ALLOW_KEYCLOAK_ADMIN_PASSWORD_FALLBACK=${ALLOW_KEYCLOAK_ADMIN_PASSWORD_FALLBACK:-true}
      # 098 Slack identity linking HMAC secret (shared with slack-bot)
      - SLACK_LINK_HMAC_SECRET=${SLACK_LINK_HMAC_SECRET:-${SLACK_SIGNING_SECRET:-}}
      - SLACK_BOT_TOKEN=${SLACK_BOT_TOKEN:-}
      # Canonical Slack workspace namespace used by UI-managed Slack ReBAC/routes.
      - SLACK_WORKSPACE_ALIAS=${SLACK_WORKSPACE_ALIAS:-CAIPE}
      # UI BFF -> Slack bot admin API. The browser never calls the bot directly;
      # caipe-ui obtains this token server-side using OIDC_CLIENT_ID/_SECRET.
      - SLACK_BOT_ADMIN_URL=${SLACK_BOT_ADMIN_URL:-http://slack-bot:3001}
      - SLACK_BOT_ADMIN_AUDIENCE=${SLACK_BOT_ADMIN_AUDIENCE:-caipe-slack-bot-admin}
      - SLACK_BOT_ADMIN_TOKEN_URL=${SLACK_BOT_ADMIN_TOKEN_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/token}
      - SLACK_BOT_ADMIN_DEV_AUTH_ENABLED=${SLACK_BOT_ADMIN_DEV_AUTH_ENABLED:-false}
      - SLACK_BOT_ADMIN_DEV_TOKEN=${SLACK_BOT_ADMIN_DEV_TOKEN:-}
      - WEBEX_INTEGRATION_BOT_ACCESS_TOKEN=${WEBEX_INTEGRATION_BOT_ACCESS_TOKEN:-}
      - WEBEX_WORKSPACE_ALIAS=${WEBEX_WORKSPACE_ALIAS:-CAIPE-WEBEX}
      - WEBEX_LINK_HMAC_SECRET=${WEBEX_LINK_HMAC_SECRET:-${WEBEX_SIGNING_SECRET:-}}
      - WEBEX_BOT_ADMIN_URL=${WEBEX_BOT_ADMIN_URL:-http://webex-bot:3002}
      - WEBEX_BOT_ADMIN_CLIENT_ID=${WEBEX_BOT_ADMIN_CLIENT_ID:-caipe-ui}
      - WEBEX_BOT_ADMIN_CLIENT_SECRET=${WEBEX_BOT_ADMIN_CLIENT_SECRET:-}
      - WEBEX_BOT_ADMIN_AUDIENCE=${WEBEX_BOT_ADMIN_AUDIENCE:-caipe-webex-bot-admin}
      - WEBEX_BOT_ADMIN_TOKEN_URL=${WEBEX_BOT_ADMIN_TOKEN_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/token}
      # App config: loaded at Next.js startup via instrumentation.ts.
      # APP_CONFIG_FILE on the host (set in .env) selects which file is
      # mounted — defaults to the committed example template (./config/app-config.yaml,
      # which seeds nothing). For local dev, set in .env:
      #   APP_CONFIG_FILE=./config/app-config.local.yaml
      # config/*.local.yaml is gitignored (see .gitignore).
      - APP_CONFIG_PATH=/app/config/app-config.yaml
    volumes:
      # Override host path with CAIPE_APP_CONFIG_FILE (preferred; matches main's naming) or
      # legacy APP_CONFIG_FILE (kept for back-compat with existing .env files on this branch).
      # If neither is set, mount the committed example template config/app-config.yaml.
      - ${CAIPE_APP_CONFIG_FILE:-${APP_CONFIG_FILE:-./config/app-config.yaml}}:/app/config/app-config.yaml:ro
      - ./charts/ai-platform-engineering/data/skills:/charts/ai-platform-engineering/data/skills:ro
      # Hot-reload bind mounts. These are SAFE in prod because the
      # `runner` stage uses Next standalone output and never reads
      # /app/src, /app/next.config.ts, etc. They are USED in dev when
      # the `dev` stage runs `next dev` (CAIPE_UI_BUILD_TARGET=dev).
      # Note: ui/public is intentionally NOT mounted — mounting it
      # would shadow the prod runner's prebuilt public/ (entrypoint.sh
      # writes env-config.js into it). For public/ asset edits in hot
      # mode, just rebuild the image; src/ changes are the 99% case.
      - ./ui/src:/app/src
      - ./ui/next.config.ts:/app/next.config.ts
      - ./ui/tsconfig.json:/app/tsconfig.json
      - ./ui/postcss.config.mjs:/app/postcss.config.mjs
      - ./ui/eslint.config.mjs:/app/eslint.config.mjs
      # Do not mount /app/node_modules or /app/.next here: this compose file
      # bind-mounts only selected source/config files, so the dev image's
      # installed dependencies and generated Next cache stay intact.
    # Hot-reload command: next dev rebuilds on edits to bind-mounted ./ui/src.
    command: npm run dev
    depends_on:
      caipe-supervisor:
        condition: service_started
        required: false
      # RBAC profile: wait for realm/JWKS readiness before UI startup seed/scope sync
      # calls Keycloak Admin APIs. Non-RBAC UI runs keep working because this
      # dependency is optional when the keycloak service is not selected.
      keycloak:
        condition: service_healthy
        required: false
    profiles:
      - caipe-ui

  # caipe-ui-prod — prod-parity sibling of caipe-ui (see header above).
  # Runs the runner stage with NODE_ENV=production and `next start` against
  # the prebuilt .next/standalone bundle. No source bind-mounts: source edits
  # require a rebuild (`make caipe-ui-prod`). Same env/secrets/depends_on as
  # caipe-ui — only build target, NODE_ENV, image suffix, command, and
  # profile differ. Env-var defaults are duplicated rather than anchored
  # because docker-compose YAML anchors don't merge `environment:` lists
  # cleanly across services.
  caipe-ui-prod:
    build:
      context: .
      dockerfile: build/Dockerfile.caipe-ui
      target: runner
      args:
        - IMAGE_TAG=${IMAGE_TAG:-localtag}
        - GIT_COMMIT=${GIT_COMMIT:-unknown}
        - BUILD_DATE=${BUILD_DATE:-}
        # Local prod-parity rebuilds skip Next's duplicate typecheck; run
        # `npm run lint` / targeted tests separately when validating changes.
        - CAIPE_UI_FAST_BUILD=${CAIPE_UI_FAST_BUILD:-true}
    image: ghcr.io/cnoe-io/caipe-ui:${IMAGE_TAG:-localtag}-prod
    container_name: caipe-ui-prod
    # Same host port as caipe-ui — these two services are mutually exclusive.
    # Keycloak's caipe-ui client only allow-lists http://localhost:3000/*.
    ports: ["3000:3000"]
    # Slack-bot, webex-bot, dynamic-agents, and a handful of other sibling
    # services hardcode the in-network hostname `http://caipe-ui:3000` for
    # CAIPE_API_URL/CREDENTIAL_API_URL. The hot-reload `caipe-ui` service
    # gets that hostname for free via its container_name, but this prod
    # sibling needs an explicit network alias so the same env var resolves
    # when only `caipe-ui-prod` is up.
    networks:
      default:
        aliases:
          - caipe-ui
    env_file:
      - .env
    environment:
      - NODE_ENV=production
      - E2E_RUN=${E2E_RUN:-false}
      - A2A_BASE_URL=${A2A_BASE_URL:-http://caipe-supervisor:8000}
      - RAG_ENABLED=${RAG_ENABLED:-true}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - RAG_URL=${RAG_URL:-http://localhost:9446}
      - NEXTAUTH_SECRET=${NEXTAUTH_SECRET:-caipe-dev-secret-change-in-production}
      - NEXTAUTH_URL=${NEXTAUTH_URL:-http://localhost:3000}
      - SSO_ENABLED=${SSO_ENABLED:-false}
      - OIDC_CLIENT_ID=${OIDC_CLIENT_ID:-}
      - OIDC_CLIENT_SECRET=${OIDC_CLIENT_SECRET:-}
      - OIDC_ISSUER=${OIDC_ISSUER:-}
      - OIDC_EXTRA_AUDIENCES=${OIDC_EXTRA_AUDIENCES:-agentgateway}
      - OIDC_ACCEPTED_AUDIENCES=${OIDC_ACCEPTED_AUDIENCES:-caipe-platform}
      - OPENFGA_RECONCILE_ENABLED=${OPENFGA_RECONCILE_ENABLED:-false}
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      # Keep local .env role defaults from affecting RAG authorization paths.
      - ALLOW_TRUSTED_NETWORK=false
      - TRUSTED_NETWORK_CIDRS=
      - TRUSTED_NETWORK_DEFAULT_ROLE=
      - RBAC_ADMIN_GROUPS=
      - RBAC_DEFAULT_AUTHENTICATED_ROLE=
      - RBAC_READONLY_GROUPS=
      - RBAC_INGESTONLY_GROUPS=
      - AUTHZ_TRACING_ENABLED=${AUTHZ_TRACING_ENABLED:-false}
      - OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-}
      - OIDC_DISCOVERY_URL=${OIDC_DISCOVERY_URL:-http://keycloak:7080/realms/caipe}
      - OIDC_GROUP_CLAIM=${OIDC_GROUP_CLAIM:-}
      - OIDC_REQUIRED_GROUP=${OIDC_REQUIRED_GROUP:-}
      - OIDC_REQUIRED_ADMIN_GROUP=${OIDC_REQUIRED_ADMIN_GROUP:-}
      - BOOTSTRAP_ADMIN_EMAILS=${BOOTSTRAP_ADMIN_EMAILS:-}
      - IDENTITY_SYNC_LOGIN_CLAIMS_ENABLED=${IDENTITY_SYNC_LOGIN_CLAIMS_ENABLED:-true}
      - IDENTITY_SYNC_LOGIN_AUTO_CREATE_TEAMS=${IDENTITY_SYNC_LOGIN_AUTO_CREATE_TEAMS:-false}
      - CAIPE_ORG_KEY=${CAIPE_ORG_KEY:-caipe}
      - CAIPE_ORG_DISPLAY_NAME=${CAIPE_ORG_DISPLAY_NAME:-CAIPE}
      - OIDC_ENABLE_REFRESH_TOKEN=${OIDC_ENABLE_REFRESH_TOKEN:-true}
      - APP_NAME=${APP_NAME:-}
      - TAGLINE=${TAGLINE:-}
      - DESCRIPTION=${DESCRIPTION:-}
      - LOGO_URL=${LOGO_URL:-}
      - LOGO_STYLE=${LOGO_STYLE:-}
      - SPINNER_COLOR=${SPINNER_COLOR:-}
      - SHOW_POWERED_BY=${SHOW_POWERED_BY:-}
      - SUPPORT_EMAIL=${SUPPORT_EMAIL:-}
      - PREVIEW_MODE=${PREVIEW_MODE:-}
      - WORKFLOW_RUNNER_ENABLED=${WORKFLOW_RUNNER_ENABLED:-false}
      - WORKFLOWS_ENABLED=${WORKFLOWS_ENABLED:-false}
      - DYNAMIC_AGENTS_ENABLED=${DYNAMIC_AGENTS_ENABLED:-false}
      - DYNAMIC_AGENTS_URL=${DYNAMIC_AGENTS_URL:-http://dynamic-agents:8001}
      - SKILL_SCANNER_URL=${SKILL_SCANNER_URL:-http://skill-scanner:8000}
      - GITHUB_TOKEN=${GITHUB_TOKEN:-${GITHUB_PERSONAL_ACCESS_TOKEN:-}}
      - MONGODB_URI=${MONGODB_URI:-mongodb://admin:changeme@caipe-mongodb:27017/caipe?authSource=admin}
      - MONGODB_DATABASE=${MONGODB_DATABASE:-caipe}
      - CAIPE_CREDENTIALS_ENABLED=${CAIPE_CREDENTIALS_ENABLED:-false}
      - CREDENTIAL_STORE_BACKEND=${CREDENTIAL_STORE_BACKEND:-mongodb-envelope}
      - CREDENTIAL_KEY_PROVIDER=${CREDENTIAL_KEY_PROVIDER:-local-cmk}
      - CREDENTIAL_KMS_CMK_ID=${CREDENTIAL_KMS_CMK_ID:-}
      - CREDENTIAL_KMS_REGION=${CREDENTIAL_KMS_REGION:-}
      # DEV-ONLY escape hatch (prod-parity image runs NODE_ENV=production). Lets
      # the local key wrapper run despite production. INSECURE — never set true
      # in a real production deployment. See build/Dockerfile.caipe-ui header.
      - CREDENTIAL_ALLOW_INSECURE_LOCAL_KEY_WRAP=${CREDENTIAL_ALLOW_INSECURE_LOCAL_KEY_WRAP:-false}
      - CREDENTIAL_SERVICE_AUDIENCE=${CREDENTIAL_SERVICE_AUDIENCE:-caipe-credential-service}
      - CREDENTIAL_BOOTSTRAP_OAUTH_CONNECTORS=${CREDENTIAL_BOOTSTRAP_OAUTH_CONNECTORS:-false}
      - GITHUB_CLIENT_ID=${GITHUB_CLIENT_ID:-}
      - GITHUB_CLIENT_SECRET=${GITHUB_CLIENT_SECRET:-}
      - GITHUB_REDIRECT_URI=${GITHUB_REDIRECT_URI:-}
      - CONFLUENCE_CLIENT_ID=${CONFLUENCE_CLIENT_ID:-}
      - CONFLUENCE_CLIENT_SECRET=${CONFLUENCE_CLIENT_SECRET:-}
      - CONFLUENCE_REDIRECT_URI=${CONFLUENCE_REDIRECT_URI:-}
      - WEBEX_CLIENT_ID=${WEBEX_CLIENT_ID:-}
      - WEBEX_CLIENT_SECRET=${WEBEX_CLIENT_SECRET:-}
      - WEBEX_REDIRECT_URI=${WEBEX_REDIRECT_URI:-}
      - PAGERDUTY_CLIENT_ID=${PAGERDUTY_CLIENT_ID:-}
      - PAGERDUTY_CLIENT_SECRET=${PAGERDUTY_CLIENT_SECRET:-}
      - PAGERDUTY_REDIRECT_URI=${PAGERDUTY_REDIRECT_URI:-}
      - PAGERDUTY_SCOPES=${PAGERDUTY_SCOPES:-}
      - KEYCLOAK_URL=${KEYCLOAK_URL:-http://keycloak:7080}
      - KEYCLOAK_REALM=${KEYCLOAK_REALM:-caipe}
      - KEYCLOAK_RESOURCE_SERVER_ID=${KEYCLOAK_RESOURCE_SERVER_ID:-caipe-platform}
      - KEYCLOAK_CLIENT_SECRET=${KEYCLOAK_CLIENT_SECRET:-caipe-platform-dev-secret}
      - RBAC_CACHE_TTL_SECONDS=${RBAC_CACHE_TTL_SECONDS:-60}
      - KEYCLOAK_ADMIN_CLIENT_ID=${KEYCLOAK_ADMIN_CLIENT_ID:-admin-cli}
      - KEYCLOAK_ADMIN_CLIENT_SECRET=${KEYCLOAK_ADMIN_CLIENT_SECRET:-}
      - ALLOW_KEYCLOAK_ADMIN_PASSWORD_FALLBACK=${ALLOW_KEYCLOAK_ADMIN_PASSWORD_FALLBACK:-true}
      - SLACK_LINK_HMAC_SECRET=${SLACK_LINK_HMAC_SECRET:-${SLACK_SIGNING_SECRET:-}}
      - SLACK_BOT_TOKEN=${SLACK_BOT_TOKEN:-}
      - SLACK_WORKSPACE_ALIAS=${SLACK_WORKSPACE_ALIAS:-CAIPE}
      - SLACK_BOT_ADMIN_URL=${SLACK_BOT_ADMIN_URL:-http://slack-bot:3001}
      - SLACK_BOT_ADMIN_AUDIENCE=${SLACK_BOT_ADMIN_AUDIENCE:-caipe-slack-bot-admin}
      - SLACK_BOT_ADMIN_TOKEN_URL=${SLACK_BOT_ADMIN_TOKEN_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/token}
      - SLACK_BOT_ADMIN_DEV_AUTH_ENABLED=${SLACK_BOT_ADMIN_DEV_AUTH_ENABLED:-false}
      - SLACK_BOT_ADMIN_DEV_TOKEN=${SLACK_BOT_ADMIN_DEV_TOKEN:-}
      - WEBEX_INTEGRATION_BOT_ACCESS_TOKEN=${WEBEX_INTEGRATION_BOT_ACCESS_TOKEN:-}
      - WEBEX_WORKSPACE_ALIAS=${WEBEX_WORKSPACE_ALIAS:-CAIPE-WEBEX}
      - WEBEX_LINK_HMAC_SECRET=${WEBEX_LINK_HMAC_SECRET:-${WEBEX_SIGNING_SECRET:-}}
      - WEBEX_BOT_ADMIN_URL=${WEBEX_BOT_ADMIN_URL:-http://webex-bot:3002}
      - WEBEX_BOT_ADMIN_CLIENT_ID=${WEBEX_BOT_ADMIN_CLIENT_ID:-caipe-ui}
      - WEBEX_BOT_ADMIN_CLIENT_SECRET=${WEBEX_BOT_ADMIN_CLIENT_SECRET:-}
      - WEBEX_BOT_ADMIN_AUDIENCE=${WEBEX_BOT_ADMIN_AUDIENCE:-caipe-webex-bot-admin}
      - WEBEX_BOT_ADMIN_TOKEN_URL=${WEBEX_BOT_ADMIN_TOKEN_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/token}
      - APP_CONFIG_PATH=/app/config/app-config.yaml
    volumes:
      # App config only — NO source bind-mounts. The runner stage uses the
      # prebuilt .next/standalone bundle and ignores ./ui/src etc., so any
      # bind-mount here would either be dead weight or actively shadow
      # the prod runner's prebuilt assets (e.g. ui/public, which the
      # entrypoint writes env-config.js into).
      - ${CAIPE_APP_CONFIG_FILE:-${APP_CONFIG_FILE:-./config/app-config.yaml}}:/app/config/app-config.yaml:ro
      - ./charts/ai-platform-engineering/data/skills:/charts/ai-platform-engineering/data/skills:ro
    command: /app/entrypoint.sh
    depends_on:
      caipe-supervisor:
        condition: service_started
        required: false
      keycloak:
        condition: service_healthy
        required: false
    profiles:
      - caipe-ui-prod

  ####################################################################################################
  #                                      Dynamic Agents                                              #
  ####################################################################################################
  # Dynamic Agents - Standalone agent builder service using deepagents library
  # Creates ephemeral AI agents configured through the UI with MCP tool support
  #
  # Configuration:
  #   - Requires MongoDB for agent/MCP server configuration storage
  #   - Supports OIDC authentication (same config as CAIPE UI)
  #
  # Ports:
  #   - 8100: Dynamic Agents API (proxied via UI at /api/dynamic-agents)
  dynamic-agents:
    build:
      context: ./ai_platform_engineering/dynamic_agents
      dockerfile: build/Dockerfile
    image: ghcr.io/cnoe-io/caipe-dynamic-agents:${IMAGE_TAG:-localtag}
    container_name: dynamic-agents
    ports: ["8100:8001"]
    volumes:
      # Mount source for hot reload in development
      - ./ai_platform_engineering/dynamic_agents/src:/app/dynamic_agents/src
      - ${HOME}/.aws:/home/app/.aws:ro
      # Spec 102 — PDP-unavailable fallback (no-op in dev; see caipe-supervisor).
      - ${RBAC_FALLBACK_FILE:-/dev/null}:/etc/keycloak/realm-config-extras.json:ro
    env_file: [.env]
    environment:
      # Dev-only: DEV_HOT_RELOAD toggles the existing DEBUG-driven uvicorn reload
      # on the bind-mounted source under /app/dynamic_agents/src. Default off.
      - DEBUG=${DEV_HOT_RELOAD:-false}
      - AWS_CONFIG_FILE=/home/app/.aws/config
      - AWS_SHARED_CREDENTIALS_FILE=/home/app/.aws/credentials
      - RBAC_FALLBACK_CONFIG_PATH=${RBAC_FALLBACK_CONFIG_PATH:-}
      - AWS_BEDROCK_STREAMING=${AWS_BEDROCK_STREAMING:-false}
      - AWS_BEDROCK_USE_CONVERSE_API=${AWS_BEDROCK_USE_CONVERSE_API:-true}
      - AWS_BEDROCK_BASE_MODEL_ID=${AWS_BEDROCK_BASE_MODEL_ID:-}
      # MongoDB (same as CAIPE UI)
      - MONGODB_URI=${MONGODB_URI:-mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@caipe-mongodb:27017/${MONGODB_DATABASE:-caipe}?authSource=admin}
      - MONGODB_DATABASE=${MONGODB_DATABASE:-caipe}
      # Authentication (same OIDC config as CAIPE UI)
      - AUTH_ENABLED=${AUTH_ENABLED:-false}
      - OIDC_ISSUER=${OIDC_ISSUER:-}
      - OIDC_CLIENT_ID=${OIDC_CLIENT_ID:-}
      - OIDC_DISCOVERY_URL=${OIDC_DISCOVERY_URL:-}
      - OIDC_GROUP_CLAIM=${OIDC_GROUP_CLAIM:-}
      - OIDC_REQUIRED_ADMIN_GROUP=${OIDC_REQUIRED_ADMIN_GROUP:-}
      # Tracing (Langfuse)
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      # Runtime configuration
      - LOG_LEVEL=${DYNAMIC_AGENTS_LOG_LEVEL:-DEBUG}
      - DEFAULT_RUNTIME_BACKEND=${DEFAULT_RUNTIME_BACKEND:-store}
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - AGENT_RUNTIME_TTL_SECONDS=${AGENT_RUNTIME_TTL_SECONDS:-60}
      # Keep uvicorn reload off in Docker. DEBUG=true from .env can make the
      # mounted dev service hang during startup; the UI gateway still injects
      # dev user context when auth is disabled.
      - DEBUG=${DYNAMIC_AGENTS_DEBUG:-false}
      # CORS_ORIGINS: must be a JSON array (bare "*" from prod default is not valid JSON for list[str])
      # To restrict origins, set DYNAMIC_AGENTS_CORS_ORIGINS='["http://localhost:3000"]' in .env
      - CORS_ORIGINS=${DYNAMIC_AGENTS_CORS_ORIGINS:-["*"]}
      # Agent Gateway — MCP routing through AG for dynamic agents
      - AGENT_GATEWAY_URL=${AGENT_GATEWAY_URL:-http://agentgateway:4000}
      # AgentGateway fronts MCP targets with one route/backend per server so one
      # failing optional MCP does not break unrelated tools.
      - AGENT_GATEWAY_MCP_SERVER_IDS=${AGENT_GATEWAY_MCP_SERVER_IDS:-all}
      # Optional shared secret for signed Dynamic Agent -> AgentGateway agent_id context.
      # Leave unset to keep only the coarse user-level AgentGateway/OpenFGA gate.
      - CAIPE_AGENT_CONTEXT_HMAC_SECRET=${CAIPE_AGENT_CONTEXT_HMAC_SECRET:-}
      # Connections & Secrets credential exchange. Disabled by default; when
      # USE_IMPERSONATION_TOKENS=true, DA resolves MCP credential refs via the
      # server-to-server credential API instead of browser/session cookies.
      - CAIPE_CREDENTIALS_ENABLED=${CAIPE_CREDENTIALS_ENABLED:-false}
      - CREDENTIAL_API_URL=${CREDENTIAL_API_URL:-http://caipe-ui:3000/api/credentials}
      - CREDENTIAL_SERVICE_AUDIENCE=${CREDENTIAL_SERVICE_AUDIENCE:-caipe-credential-service}
      - USE_IMPERSONATION_TOKENS=${USE_IMPERSONATION_TOKENS:-false}
      # Static service-account fallback for shared-token MCP servers (GitHub/GitLab).
      # When a caller has NOT connected their personal provider account, DA injects
      # these org PATs on X-CAIPE-Provider-Token so those tools stay backward
      # compatible. Connected users always get their own OAuth token instead.
      - GITHUB_PERSONAL_ACCESS_TOKEN=${GITHUB_PERSONAL_ACCESS_TOKEN:-}
      - GITLAB_PERSONAL_ACCESS_TOKEN=${GITLAB_PERSONAL_ACCESS_TOKEN:-${GITLAB_TOKEN:-}}
      # Spec 102 Phase 11.4 — require Bearer JWT on every DA request.
      # The BFF (ui/src/lib/da-proxy.ts and the two probe / suggest routes)
      # now always sends Authorization: Bearer alongside the legacy
      # X-User-Context header. Set DA_REQUIRE_BEARER=false to roll back if
      # an unmigrated caller is discovered in production.
      - DA_REQUIRE_BEARER=${DA_REQUIRE_BEARER:-true}
      # OpenFGA runtime PDP — Dynamic Agents repeats the BFF can_use check
      # before start/invoke/resume so direct bearer callers cannot bypass ReBAC.
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      - OPENFGA_STORE_ID=${OPENFGA_STORE_ID:-}
      - AUTHZ_TRACING_ENABLED=${AUTHZ_TRACING_ENABLED:-false}
      - OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-}

      # Checkpoint configuration
      - LANGGRAPH_CHECKPOINT_TYPE=mongodb
      - LANGGRAPH_CHECKPOINT_MONGODB_URI=${MONGODB_URI:-mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@caipe-mongodb:27017/${MONGODB_DATABASE:-caipe}?authSource=admin}
    healthcheck:
      # Use /healthz (matches FastAPI route in routes/health.py and the Helm chart probes).
      # Parse the JSON body: a 200 response with {"status": "unhealthy"} (e.g. Mongo down)
      # must still fail the healthcheck, otherwise the container appears healthy while broken.
      test: ["CMD-SHELL", "python -c \"import json,urllib.request,sys; r=urllib.request.urlopen('http://localhost:8001/healthz',timeout=3); sys.exit(0 if r.status==200 and json.load(r).get('status')=='healthy' else 1)\""]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s
    restart: unless-stopped
    profiles:
      - dynamic-agents

  ####################################################################################################
  #                                      RAG Services                                                #
  ####################################################################################################

  rag-server:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.server
    image: ghcr.io/cnoe-io/caipe-rag-server:${IMAGE_TAG:-localtag}
    container_name: rag-server
    # Network alias preserves the legacy `rag_server` hostname for any client
    # that still has it hardcoded. AGW's hickory-dns resolver rejects DNS
    # labels with underscores (RFC 1035), so the canonical service name is
    # now `rag-server` (hyphen) and `rag_server` is a compatibility alias.
    networks:
      default:
        aliases:
          - rag_server
    # Keycloak discovery returns jwks_uri on localhost:7080; map localhost → host so fetches work in-container.
    extra_hosts:
      - "localhost:host-gateway"
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/server/src/server:/app/server/src/server
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
      - ${HOME}/.aws:/home/app/.aws:ro
      # Spec 102 — PDP-unavailable fallback (no-op in dev; see caipe-supervisor).
      - ${RBAC_FALLBACK_FILE:-/dev/null}:/etc/keycloak/realm-config-extras.json:ro
    ports: ["9446:9446"]
    environment:
      # Dev-only: DEV_HOT_RELOAD=true → uvicorn --reload watches the bind-mounted
      # source under /app/server/src/server and /app/common/src/common. Default off.
      - DEV_HOT_RELOAD=${DEV_HOT_RELOAD:-false}
      - AWS_CONFIG_FILE=/home/app/.aws/config
      - AWS_SHARED_CREDENTIALS_FILE=/home/app/.aws/credentials
      - LOG_LEVEL=${RAG_SERVER_LOG_LEVEL:-INFO}
      - RBAC_FALLBACK_CONFIG_PATH=${RBAC_FALLBACK_CONFIG_PATH:-}
      - REDIS_URL=redis://rag-redis:6379/0
      - NEO4J_ADDR=neo4j://neo4j:7687
      - NEO4J_ONTOLOGY_ADDR=neo4j://neo4j-ontology:7688
      - NEO4J_USERNAME=neo4j
      - NEO4J_PASSWORD=dummy_password
      - MILVUS_URI=http://milvus-standalone:19530
      - ONTOLOGY_AGENT_RESTAPI_ADDR=http://agent_ontology:8098
      # false = vector RAG only (no Neo4j); use graph_rag profile + ENABLE_GRAPH_RAG=true for graph
      - ENABLE_GRAPH_RAG=${ENABLE_GRAPH_RAG:-false}
      - CLEANUP_INTERVAL=86400
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
      # Embeddings configuration
      - EMBEDDINGS_PROVIDER=${EMBEDDINGS_PROVIDER:-azure-openai}
      - EMBEDDINGS_MODEL=${EMBEDDINGS_MODEL:-text-embedding-3-large}
      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-}
      # OIDC JWT validation — defaults to local Keycloak; .env overrides if set.
      # rag_server uses OIDC_DISCOVERY_URL (Docker-internal) to fetch JWKS
      # and OIDC_ISSUER / OIDC_ISSUER_URL (browser-facing) to validate the "iss" claim in JWTs.
      # OIDC_JWKS_URL: direct certs URL inside Docker network (optional; speeds JWKS fetch).
      # OIDC_AUDIENCE: resource-server audience (caipe-platform) added by Keycloak audience mapper.
      - OIDC_ISSUER=${OIDC_ISSUER:-http://localhost:7080/realms/caipe}
      - OIDC_ISSUER_URL=${OIDC_ISSUER_URL:-http://localhost:7080/realms/caipe}
      - OIDC_CLIENT_ID=${OIDC_CLIENT_ID:-caipe-ui}
      - OIDC_AUDIENCE=${OIDC_AUDIENCE:-caipe-platform}
      - OIDC_JWKS_URL=${OIDC_JWKS_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/certs}
      - OIDC_DISCOVERY_URL=${OIDC_DISCOVERY_URL:-http://keycloak:7080/realms/caipe/.well-known/openid-configuration}
      - OIDC_GROUP_CLAIM=
      # Ingestor OIDC configuration (for OAuth2 client credentials)
      - INGESTOR_OIDC_ISSUER=${INGESTOR_OIDC_ISSUER:-}
      - INGESTOR_OIDC_CLIENT_ID=${INGESTOR_OIDC_CLIENT_ID}
      # Client-credentials ingestors keep a service role; human access is OpenFGA-only.
      - RBAC_CLIENT_CREDENTIALS_ROLE=${RBAC_CLIENT_CREDENTIALS_ROLE:-ingestonly}
      # 098 team-scoped KB / datasource checks (MongoDB same as UI)
      - RBAC_TEAM_SCOPE_ENABLED=${RBAC_TEAM_SCOPE_ENABLED:-true}
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      - CAIPE_UNSAFE_RBAC_BYPASS=${CAIPE_UNSAFE_RBAC_BYPASS:-false}
      - RBAC_MONGODB_URI=${RBAC_MONGODB_URI:-mongodb://admin:changeme@caipe-mongodb:27017/caipe?authSource=admin}
      - RBAC_MONGODB_DATABASE=${RBAC_MONGODB_DATABASE:-caipe}
    restart: unless-stopped
    env_file: [.env]
    depends_on:
      rag-redis:
        condition: service_started
      milvus-standalone:
        condition: service_healthy
    profiles:
      - rag
      - graph_rag

  agent_ontology:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.agent-ontology
    image: ghcr.io/cnoe-io/caipe-rag-agent-ontology:${IMAGE_TAG:-localtag}
    container_name: agent_ontology
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/agent_ontology/src/agent_ontology:/app/agent_ontology/src/agent_ontology
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    ports: ["8098:8098"]
    environment:
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - REDIS_URL=redis://rag-redis:6379/0
      - NEO4J_ADDR=neo4j://neo4j:7687
      - NEO4J_ONTOLOGY_ADDR=neo4j://neo4j-ontology:7688
      - NEO4J_USERNAME=neo4j
      - NEO4J_PASSWORD=dummy_password
      - SYNC_INTERVAL=86400
      - ENABLE_TRACING=${ENABLE_TRACING:-false}
      - LANGFUSE_HOST=${LANGFUSE_HOST}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - LANGFUSE_SESSION_ID=${LANGFUSE_SESSION_ID}
      - LANGFUSE_USER_ID=${LANGFUSE_USER_ID}
    env_file: [.env]
    restart: unless-stopped
    depends_on: [rag-server, neo4j, rag-redis]
    profiles:
      - graph_rag

  ####################################################################################################
  #                                      RAG Ingestors                                               #
  ####################################################################################################
  web_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: web-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - PYTHONPATH=/app/common/src:/app/ingestors/src
      - INGESTOR_TYPE=webloader
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - REDIS_URL=redis://rag-redis:6379/0
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${WEBLOADER_CHECK_INTERVAL:-86400}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - SCRAPY_WORKER_POOL_SIZE=${SCRAPY_WORKER_POOL_SIZE:-3}
      - WEBLOADER_MAX_INGESTION_TASKS=${WEBLOADER_MAX_INGESTION_TASKS:-3}
      # Ingestor OIDC
      - INGESTOR_OIDC_ISSUER=${INGESTOR_OIDC_ISSUER}
      - INGESTOR_OIDC_CLIENT_ID=${INGESTOR_OIDC_CLIENT_ID}
      - INGESTOR_OIDC_CLIENT_SECRET=${INGESTOR_OIDC_CLIENT_SECRET}
      - INGESTOR_OIDC_DISCOVERY_URL=${INGESTOR_OIDC_DISCOVERY_URL:-}
      - INGESTOR_OIDC_SCOPE=${INGESTOR_OIDC_SCOPE:-}
    env_file: [.env]
    restart: unless-stopped
    profiles:
      - web-ingestor

  backstage_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: backstage-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=backstage
      - BACKSTAGE_URL=${BACKSTAGE_URL}
      - BACKSTAGE_API_TOKEN=${BACKSTAGE_API_TOKEN}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${BACKSTAGE_SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${BACKSTAGE_INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${BACKSTAGE_EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${BACKSTAGE_MAX_DOCUMENTS_PER_INGEST:-25}
      - MAX_INGESTION_CONCURRENCY=${BACKSTAGE_MAX_INGESTION_CONCURRENCY:-10}
      - IGNORE_TYPES=${BACKSTAGE_IGNORE_TYPES:-api,component,domain,group,location,system,template,user}
    env_file: [.env]
    restart: unless-stopped
    profiles:
      - backstage-ingestor

  aws_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: aws-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
      - ${HOME}/.aws/credentials:/home/app/.aws/credentials:ro
    environment:
      - INGESTOR_TYPE=aws
      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID:-}
      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY:-}
      - AWS_SESSION_TOKEN=${AWS_SESSION_TOKEN:-}
      - AWS_REGION=${AWS_REGION:-}
      - AWS_DEFAULT_REGION=${AWS_DEFAULT_REGION:-us-east-2}
      - AWS_PROFILE=${AWS_PROFILE:-}
      - AWS_DEFAULT_PROFILE=${AWS_DEFAULT_PROFILE:-}
      - RESOURCE_TYPES=${AWS_RESOURCE_TYPES:-iam:user,ec2:instance,ec2:volume,ec2:natgateway,ec2:vpc,ec2:subnet,ec2:security-group,eks:cluster,s3:bucket,elasticloadbalancing:loadbalancer,route53:hostedzone,rds:db,lambda:function,dynamodb:table}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
      - AWS_ACCOUNT_LIST=${AWS_ACCOUNT_LIST:-}
      - CROSS_ACCOUNT_ROLE_NAME=${CROSS_ACCOUNT_ROLE_NAME:-caipe-read-only}
    restart: no
    env_file: [.env]
    profiles:
      - aws-ingestor

  k8s_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: k8s-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
      - ~/.kube/config:/home/app/.kube/config:ro
    environment:
      - INGESTOR_TYPE=k8s
      - CLUSTER_NAME=${CLUSTER_NAME}
      - KUBECONFIG=${KUBECONFIG:-/home/app/.kube/config}
      - KUBE_CONTEXT=${KUBE_CONTEXT:-}
      - AWS_PROFILE=${AWS_PROFILE:-}
      - AWS_DEFAULT_PROFILE=${AWS_DEFAULT_PROFILE:-}
      - AWS_DEFAULT_REGION=${AWS_DEFAULT_REGION:-us-east-2}
      - AWS_REGION=${AWS_REGION:-}
      - IN_CLUSTER=${K8S_IN_CLUSTER:-false}
      - RESOURCE_LIST=${K8S_RESOURCE_LIST:-Certificate,ClusterIssuer,CronJob,DaemonSet,Deployment,Ingress,IngressClass,Issuer,Job,Namespace,Node,Service,StatefulSet,StorageClass}
      - IGNORE_FIELD_LIST=${K8S_IGNORE_FIELD_LIST:-metadata.annotations.kubectl.kubernetes.io,metadata.labels.app.kubernetes.io,metadata.managedFields,metadata.selfLink}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    profiles:
      - k8s-ingestor

  slack_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: slack-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=slack
      - SLACK_BOT_TOKEN=${SLACK_BOT_TOKEN}
      - SLACK_BOT_NAME=${SLACK_BOT_NAME}
      - SLACK_WORKSPACE_URL=${SLACK_WORKSPACE_URL:-https://slack.com}
      - SLACK_CHANNELS=${SLACK_CHANNELS:-{}}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    profiles:
      - slack-ingestor

  webex_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: webex-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=webex
      - WEBEX_ACCESS_TOKEN=${WEBEX_ACCESS_TOKEN}
      - WEBEX_BOT_NAME=${WEBEX_BOT_NAME}
      - WEBEX_SPACES=${WEBEX_SPACES:-{}}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-60}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    profiles:
      - webex-ingestor

  argocd_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: argocd-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=argocdv3
      - SERVER_URL=${SERVER_URL}
      - ARGOCD_AUTH_TOKEN=${ARGOCD_AUTH_TOKEN}
      - ARGOCD_VERIFY_SSL=${ARGOCD_VERIFY_SSL:-true}
      - ARGOCD_FILTER_PROJECTS=${ARGOCD_FILTER_PROJECTS:-}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    profiles:
      - argocd-ingestor

  github_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: ghcr.io/cnoe-io/caipe-rag-ingestors:${IMAGE_TAG:-localtag}
    container_name: github-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=github
      # Authentication: Use either PAT or GitHub App
      - GITHUB_TOKEN=${GITHUB_TOKEN:-}
      # GITHUB_APP_ID, GITHUB_APP_PRIVATE_KEY, GITHUB_APP_INSTALLATION_ID for GitHub App auth
      - GITHUB_ORG=${GITHUB_ORG}
      - GITHUB_ORGS=${GITHUB_ORGS:-}
      - GITHUB_APP_INSTALLATION_IDS=${GITHUB_APP_INSTALLATION_IDS:-}
      - GITHUB_API_URL=${GITHUB_API_URL:-https://api.github.com/graphql}
      - GITHUB_REST_API_URL=${GITHUB_REST_API_URL:-https://api.github.com}
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - FETCH_TEAM_DETAILS=${FETCH_TEAM_DETAILS:-true}
      - FETCH_ORG_EMAILS=${FETCH_ORG_EMAILS:-false}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    profiles:
      - github-ingestor

  jira_ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    container_name: jira-ingestor
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/ingestors/src/ingestors:/app/ingestors/src/ingestors
      - ./ai_platform_engineering/knowledge_bases/rag/common/src/common:/app/common/src/common
    environment:
      - INGESTOR_TYPE=jira
      - JIRA_URL=${JIRA_URL}
      - JIRA_EMAIL=${JIRA_EMAIL}
      - ATLASSIAN_TOKEN=${ATLASSIAN_TOKEN}
      - JIRA_PROJECTS=${JIRA_PROJECTS}
      - JIRA_PAGE_SIZE=${JIRA_PAGE_SIZE:-100}
      - RAG_SERVER_URL=http://rag-server:9446
      - SYNC_INTERVAL=${SYNC_INTERVAL:-86400}
      - INIT_DELAY_SECONDS=${INIT_DELAY_SECONDS:-0}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - EXIT_AFTER_FIRST_SYNC=${EXIT_AFTER_FIRST_SYNC:-false}
      - MAX_DOCUMENTS_PER_INGEST=${MAX_DOCUMENTS_PER_INGEST:-1000}
    restart: no
    env_file: [.env]
    depends_on: [rag-server]
    profiles:
      - jira-ingestor

  # Dummy Graph Ingestor - for testing graph RAG with sample data
  # Requires RAG services to be running first: docker compose -f docker-compose.dev.yaml --profile graph_rag up -d
  # Then run: docker compose -f docker-compose.dev.yaml --profile dummy-ingestor up --build dummy-structured-ingestor
  dummy-structured-ingestor:
    build:
      context: ./ai_platform_engineering/knowledge_bases/rag
      dockerfile: ./build/Dockerfile.ingestors
    image: dummy-structured-ingestor:local
    container_name: dummy-ingestor
    environment:
      - RAG_SERVER_URL=${RAG_SERVER_URL:-http://rag-server:9446}
      - INGESTOR_TYPE=dummy_structured_ingestor
      - DUMMY_ENTITIES_FILE=/app/data/dummy_entities.json
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
    volumes:
      - ./ai_platform_engineering/knowledge_bases/rag/tests/rag_graph_test:/app/data:ro
    profiles:
      - dummy-ingestor

  # RAG Dependencies
  neo4j:
    image: neo4j:latest
    container_name: neo4j
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/neo4j/logs:/logs
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/neo4j/config:/config
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/neo4j/data:/data
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/neo4j/plugins:/plugins
    ports: ["7474:7474", "7687:7687"]
    restart: unless-stopped
    environment:
      - NEO4J_AUTH=neo4j/dummy_password
      - NEO4J_PLUGINS='["apoc"]'
      - NEO4J_apoc_export_file_enabled=true
      - NEO4J_apoc_import_file_enabled=true
      - NEO4J_apoc_import_file_use__neo4j__config=true
    profiles:
      - graph_rag
      - deps

  rag-redis:
    image: redis
    container_name: rag-redis
    volumes:
      - rag_redis_data:/data
    command: ["/bin/sh", "-c", "redis-server --save 60 1 --appendonly yes"]
    ports: [":6379"]
    restart: unless-stopped
    profiles:
      - rag
      - graph_rag

  caipe-redis:
    image: redis
    container_name: caipe-redis
    command:
      - /bin/sh
      - -c
      - redis-server
    ports:
      - "6379:6379"
    restart: unless-stopped
    profiles:
      - deps

  milvus-standalone:
    image: milvusdb/milvus:v2.6.0
    container_name: milvus-standalone
    command: ["milvus", "run", "standalone"]
    environment:
      - MINIO_REGION=us-east-1
      - ETCD_ENDPOINTS=etcd:2379
      - MINIO_ADDRESS=milvus-minio:9000
      - LOG_LEVEL=error
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/milvus:/var/lib/milvus
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9091/healthz"]
      interval: 30s
      start_period: 90s
      timeout: 20s
      retries: 3
    ports: ["19530:19530", "9091:9091"]
    depends_on: [etcd, milvus-minio]
    profiles:
      - rag
      - graph_rag
      - deps

  etcd:
    image: quay.io/coreos/etcd:v3.5.18
    container_name: milvus-etcd
    environment:
      - ETCD_AUTO_COMPACTION_MODE=revision
      - ETCD_AUTO_COMPACTION_RETENTION=1000
      - ETCD_QUOTA_BACKEND_BYTES=4294967296
      - ETCD_SNAPSHOT_COUNT=50000
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/etcd:/etcd
    command: etcd -advertise-client-urls=http://etcd:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd
    healthcheck:
      test: ["CMD", "etcdctl", "endpoint", "health"]
      interval: 30s
      timeout: 20s
      retries: 3
    profiles:
      - rag
      - graph_rag
      - deps

  milvus-minio:
    image: minio/minio:RELEASE.2024-05-28T17-19-04Z
    container_name: milvus-minio
    environment:
      - MINIO_ACCESS_KEY=minioadmin
      - MINIO_SECRET_KEY=minioadmin
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/minio:/minio_data
    command: minio server /minio_data --console-address ":9001"
    healthcheck:
      test: ["CMD", "mc", "ready", "local"]
      interval: 30s
      timeout: 20s
      retries: 3
    profiles:
      - rag
      - graph_rag
      - deps

  # Langfuse Tracing Services
  langfuse-worker:
    image: langfuse/langfuse-worker:3
    container_name: langfuse-worker
    restart: always
    depends_on:
      langfuse-postgres:
        condition: service_healthy
      langfuse-minio:
        condition: service_healthy
      langfuse-redis:
        condition: service_healthy
      langfuse-clickhouse:
        condition: service_healthy
    ports: ["127.0.0.1:3030:3030"]
    environment:
      - DATABASE_URL=postgresql://postgres:postgres@langfuse-postgres:5432/postgres
      - SALT=mysalt
      - ENCRYPTION_KEY=0000000000000000000000000000000000000000000000000000000000000000
      - CLICKHOUSE_MIGRATION_URL=clickhouse://langfuse-clickhouse:9000
      - CLICKHOUSE_URL=http://langfuse-clickhouse:8123
      - CLICKHOUSE_USER=clickhouse
      - CLICKHOUSE_PASSWORD=clickhouse
      - CLICKHOUSE_CLUSTER_ENABLED=false
      - LANGFUSE_S3_EVENT_UPLOAD_BUCKET=langfuse
      - LANGFUSE_S3_EVENT_UPLOAD_REGION=us-east-1
      - LANGFUSE_S3_EVENT_UPLOAD_ACCESS_KEY_ID=minio
      - LANGFUSE_S3_EVENT_UPLOAD_SECRET_ACCESS_KEY=miniosecret
      - LANGFUSE_S3_EVENT_UPLOAD_ENDPOINT=http://langfuse-minio:9000
      - LANGFUSE_S3_EVENT_UPLOAD_FORCE_PATH_STYLE=true
      - LANGFUSE_S3_EVENT_UPLOAD_PREFIX=events/
      - LANGFUSE_S3_MEDIA_UPLOAD_BUCKET=langfuse
      - LANGFUSE_S3_MEDIA_UPLOAD_REGION=us-east-1
      - LANGFUSE_S3_MEDIA_UPLOAD_ACCESS_KEY_ID=minio
      - LANGFUSE_S3_MEDIA_UPLOAD_SECRET_ACCESS_KEY=miniosecret
      - LANGFUSE_S3_MEDIA_UPLOAD_ENDPOINT=http://langfuse-minio:9000
      - LANGFUSE_S3_MEDIA_UPLOAD_FORCE_PATH_STYLE=true
      - LANGFUSE_S3_MEDIA_UPLOAD_PREFIX=media/
      - REDIS_HOST=langfuse-redis
      - REDIS_AUTH=myredissecret
    profiles:
      - tracing

  langfuse-web:
    image: langfuse/langfuse:3
    container_name: langfuse-web
    restart: always
    depends_on:
      langfuse-postgres:
        condition: service_healthy
      langfuse-minio:
        condition: service_healthy
      langfuse-redis:
        condition: service_healthy
      langfuse-clickhouse:
        condition: service_healthy
    ports: ["3001:3000"]
    environment:
      - DATABASE_URL=postgresql://postgres:postgres@langfuse-postgres:5432/postgres
      - SALT=mysalt
      - ENCRYPTION_KEY=0000000000000000000000000000000000000000000000000000000000000000
      - CLICKHOUSE_MIGRATION_URL=clickhouse://langfuse-clickhouse:9000
      - CLICKHOUSE_URL=http://langfuse-clickhouse:8123
      - CLICKHOUSE_USER=clickhouse
      - HOSTNAME=0.0.0.0
      - CLICKHOUSE_PASSWORD=clickhouse
      - CLICKHOUSE_CLUSTER_ENABLED=false
      - LANGFUSE_S3_EVENT_UPLOAD_BUCKET=langfuse
      - LANGFUSE_S3_EVENT_UPLOAD_REGION=us-east-1
      - LANGFUSE_S3_EVENT_UPLOAD_ACCESS_KEY_ID=minio
      - LANGFUSE_S3_EVENT_UPLOAD_SECRET_ACCESS_KEY=miniosecret
      - LANGFUSE_S3_EVENT_UPLOAD_ENDPOINT=http://langfuse-minio:9000
      - LANGFUSE_S3_EVENT_UPLOAD_FORCE_PATH_STYLE=true
      - LANGFUSE_S3_EVENT_UPLOAD_PREFIX=events/
      - LANGFUSE_S3_MEDIA_UPLOAD_BUCKET=langfuse
      - LANGFUSE_S3_MEDIA_UPLOAD_REGION=us-east-1
      - LANGFUSE_S3_MEDIA_UPLOAD_ACCESS_KEY_ID=minio
      - LANGFUSE_S3_MEDIA_UPLOAD_SECRET_ACCESS_KEY=miniosecret
      - LANGFUSE_S3_MEDIA_UPLOAD_ENDPOINT=http://langfuse-minio:9000
      - LANGFUSE_S3_MEDIA_UPLOAD_FORCE_PATH_STYLE=true
      - LANGFUSE_S3_MEDIA_UPLOAD_PREFIX=media/
      - REDIS_HOST=langfuse-redis
      - REDIS_AUTH=myredissecret
      - NEXTAUTH_URL=http://localhost:3001
      - NEXTAUTH_SECRET=mysecret
    profiles:
      - tracing

  langfuse-clickhouse:
    image: clickhouse/clickhouse-server
    container_name: langfuse-clickhouse
    restart: always
    user: "101:101"
    environment:
      - CLICKHOUSE_DB=default
      - CLICKHOUSE_USER=clickhouse
      - CLICKHOUSE_PASSWORD=clickhouse
    volumes:
      - langfuse_clickhouse_data:/var/lib/clickhouse
      - langfuse_clickhouse_logs:/var/log/clickhouse-server
    ports: ["127.0.0.1:8123:8123", "127.0.0.1:9000:9000"]
    healthcheck:
      test: wget --no-verbose --tries=1 --spider http://localhost:8123/ping || exit 1
      interval: 5s
      timeout: 5s
      retries: 10
      start_period: 1s
    profiles:
      - tracing

  langfuse-minio:
    image: minio/minio
    container_name: langfuse-minio
    restart: always
    entrypoint: sh
    command: -c 'mkdir -p /data/langfuse && minio server --address ":9000" --console-address ":9001" /data'
    environment:
      - MINIO_ROOT_USER=minio
      - MINIO_ROOT_PASSWORD=miniosecret
    ports: ["9090:9000", "127.0.0.1:9092:9001"]
    volumes:
      - langfuse_minio_data:/data
    healthcheck:
      test: ["CMD", "mc", "ready", "local"]
      interval: 1s
      timeout: 5s
      retries: 5
      start_period: 1s
    profiles:
      - tracing

  langfuse-redis:
    image: redis:7
    container_name: langfuse-redis
    restart: always
    volumes:
      - langfuse_redis_data:/data
    command: --requirepass ${REDIS_AUTH:-myredissecret} --appendonly yes
    ports: ["127.0.0.1:6379:6379"]
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 3s
      timeout: 10s
      retries: 10
    profiles:
      - tracing

  langfuse-postgres:
    image: postgres:15
    container_name: langfuse-postgres
    restart: always
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 3s
      timeout: 3s
      retries: 10
    environment:
      - POSTGRES_USER=postgres
      - POSTGRES_PASSWORD=postgres
      - POSTGRES_DB=postgres
    ports: ["127.0.0.1:5432:5432"]
    volumes:
      - langfuse_postgres_data:/var/lib/postgresql/data
    profiles:
      - tracing

  # Evaluation Webhook Service
  evaluation-webhook:
    build:
      context: ./evals
      dockerfile: Dockerfile
    container_name: evaluation-webhook
    restart: unless-stopped
    depends_on:
      langfuse-web:
        condition: service_started
    ports: ["8024:8000"]
    env_file: [.env]
    environment:
      - PLATFORM_ENGINEER_URL=http://platform-engineering:8000
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://langfuse-web:3000}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY}
      - OTEL_EXPORTER_OTLP_TIMEOUT=${OTEL_EXPORTER_OTLP_TIMEOUT:-30000}
      # Skill payloads are scrubbed in-process by
      # `ai_platform_engineering.utils.tracing.install_skill_content_scrubber`
      # (attached to the TracerProvider on agent boot). That keeps
      # SKILL.md bodies + ancillary file contents out of every span
      # without redacting normal chat / tool I/O. Set
      # SKILL_TRACE_SCRUB_ENABLED=false to bypass for one debug
      # session.
      - SKILL_TRACE_SCRUB_ENABLED=${SKILL_TRACE_SCRUB_ENABLED:-true}
      # Defensive per-attribute cap (256 KiB) so one oversized
      # attribute can't push the OTel batch past Langfuse's 1 MiB
      # nginx body limit. Set to 0 to disable.
      - SKILL_TRACE_MAX_ATTR_BYTES=${SKILL_TRACE_MAX_ATTR_BYTES:-262144}
      - OPENAI_API_KEY=${OPENAI_API_KEY}
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
    volumes:
      - ./evals/datasets:/app/datasets
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    profiles:
      - evaluation

  ####################################################################################################
  #                                 Slack Bot Integration                                            #
  ####################################################################################################
  slack-bot:
    build:
      context: .
      dockerfile: build/Dockerfile.slack-bot
    image: ghcr.io/cnoe-io/caipe-slack-bot:${IMAGE_TAG:-localtag}
    container_name: slack-bot
    volumes:
      - ./ai_platform_engineering/integrations/slack_bot:/app
      - /app/.venv
      # Override host path with CAIPE_SLACK_BOT_CONFIG_FILE (preferred; matches main's naming).
      # If unset, mount the committed example template config/slack-bot-config.yaml.
      - ${CAIPE_SLACK_BOT_CONFIG_FILE:-./config/slack-bot-config.yaml}:/etc/caipe/bot-config.yaml:ro
    # DEV_HOT_RELOAD=true → install dev deps (watchfiles) and wrap python -m app
    # in `watchfiles --filter python` so source edits trigger an automatic restart
    # (Bolt is not ASGI, so uvicorn --reload is not applicable).
    command:
      - sh
      - -c
      - |
        if [ "${DEV_HOT_RELOAD:-false}" = "true" ]; then
          uv sync && exec uv run watchfiles --filter python 'python -m app' /app
        else
          uv sync --no-dev && exec python -m app
        fi
    env_file: [.env]
    ports: ["8030:3000"]  # Only used in HTTP mode (SLACK_INTEGRATION_BOT_MODE=http)
    environment:
      # Force unbuffered stdio so loguru/print output reaches `docker logs`
      # in real time. Without this, Python block-buffers stderr when the fd
      # is a pipe (docker captures aren't a TTY), which means INFO lines
      # like "RBAC denied request for slack_user=..." don't appear until a
      # buffer fills — making live debugging of denials nearly impossible.
      - PYTHONUNBUFFERED=1
      - DEV_HOT_RELOAD=${DEV_HOT_RELOAD:-false}
      - SLACK_INTEGRATION_APP_NAME=${SLACK_INTEGRATION_APP_NAME:-CAIPE}
      - SLACK_INTEGRATION_BOT_TOKEN=${SLACK_INTEGRATION_BOT_TOKEN:-${SLACK_BOT_TOKEN}}
      - SLACK_INTEGRATION_APP_TOKEN=${SLACK_INTEGRATION_APP_TOKEN:-${SLACK_APP_TOKEN}}
      - SLACK_INTEGRATION_SIGNING_SECRET=${SLACK_INTEGRATION_SIGNING_SECRET:-}
      - SLACK_INTEGRATION_BOT_MODE=${SLACK_INTEGRATION_BOT_MODE:-socket}
      - SLACK_INTEGRATION_BOT_CONFIG=${SLACK_INTEGRATION_BOT_CONFIG:-/etc/caipe/bot-config.yaml}
      - SLACK_INTEGRATION_ENABLE_AUTH=${SLACK_INTEGRATION_ENABLE_AUTH:-false}
      - SLACK_INTEGRATION_AUTH_TOKEN_URL=${SLACK_INTEGRATION_AUTH_TOKEN_URL:-}
      - SLACK_INTEGRATION_AUTH_CLIENT_ID=${SLACK_INTEGRATION_AUTH_CLIENT_ID:-}
      - SLACK_INTEGRATION_AUTH_CLIENT_SECRET=${SLACK_INTEGRATION_AUTH_CLIENT_SECRET:-}
      - SLACK_INTEGRATION_AUTH_SCOPE=${SLACK_INTEGRATION_AUTH_SCOPE:-}
      - SLACK_INTEGRATION_AUTH_AUDIENCE=${SLACK_INTEGRATION_AUTH_AUDIENCE:-}
      - CAIPE_API_URL=${CAIPE_API_URL:-http://caipe-ui:3000}
      - MONGODB_URI=mongodb://admin:changeme@caipe-mongodb:27017
      - MONGODB_DATABASE=caipe
      - SLACK_WORKSPACE_URL=${SLACK_WORKSPACE_URL:-}
      # Canonical Slack workspace namespace; incoming Slack team_id values map to this alias.
      - SLACK_WORKSPACE_ALIAS=${SLACK_WORKSPACE_ALIAS:-CAIPE}
      # 098 Enterprise RBAC (requires --profile rbac)
      - SLACK_RBAC_ENABLED=${SLACK_RBAC_ENABLED:-true}
      # Prefer UI-managed Slack channel routes, falling back to static bot config.
      - SLACK_AGENT_ROUTES_MODE=${SLACK_AGENT_ROUTES_MODE:-db_prefer}
      # Opt-in runtime onboarding for Slack channels that have no team/agent mapping yet.
      - SLACK_AUTO_ASSIGN_UNMAPPED_CHANNELS=${SLACK_AUTO_ASSIGN_UNMAPPED_CHANNELS:-false}
      - SLACK_DEFAULT_TEAM_SLUG=${SLACK_DEFAULT_TEAM_SLUG:-}
      - SLACK_DEFAULT_AGENT_ID=${SLACK_DEFAULT_AGENT_ID:-}
      # Internal admin API used by the Web UI BFF for route status/reload/config sync.
      - SLACK_ADMIN_API_ENABLED=${SLACK_ADMIN_API_ENABLED:-false}
      - SLACK_ADMIN_API_HOST=${SLACK_ADMIN_API_HOST:-0.0.0.0}
      - SLACK_ADMIN_API_PORT=${SLACK_ADMIN_API_PORT:-3001}
      - SLACK_ADMIN_DEV_AUTH_ENABLED=${SLACK_ADMIN_DEV_AUTH_ENABLED:-false}
      - SLACK_ADMIN_DEV_TOKEN=${SLACK_ADMIN_DEV_TOKEN:-}
      - SLACK_ADMIN_JWT_ISSUER=${SLACK_ADMIN_JWT_ISSUER:-http://localhost:7080/realms/caipe}
      - SLACK_ADMIN_JWKS_URL=${SLACK_ADMIN_JWKS_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/certs}
      - SLACK_ADMIN_JWT_AUDIENCE=${SLACK_ADMIN_JWT_AUDIENCE:-caipe-slack-bot-admin}
      - SLACK_ADMIN_ALLOWED_CLIENT_IDS=${SLACK_ADMIN_ALLOWED_CLIENT_IDS:-caipe-ui}
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      - OPENFGA_STORE_ID=${OPENFGA_STORE_ID:-}
      - KEYCLOAK_URL=${KEYCLOAK_URL:-http://keycloak:7080}
      - KEYCLOAK_REALM=${KEYCLOAK_REALM:-caipe}
      # Slack bot's Admin REST API credentials (user lookup by slack_user_id).
      # Distinct from the UI BFF's KEYCLOAK_ADMIN_* vars on purpose — sharing
      # those names caused a 401 collision (see ADR / architecture.md §auth).
      # The KEYCLOAK_SLACK_BOT_ADMIN_* prefix is also surface-specific so future
      # surfaces (Webex, Teams, …) get their own KEYCLOAK_<surface>_BOT_ADMIN_*.
      # Must point to a confidential client with view-users + query-users on
      # realm-management. The realm seeder grants those to caipe-platform.
      - KEYCLOAK_SLACK_BOT_ADMIN_CLIENT_ID=${KEYCLOAK_SLACK_BOT_ADMIN_CLIENT_ID:-caipe-platform}
      - KEYCLOAK_SLACK_BOT_ADMIN_CLIENT_SECRET=${KEYCLOAK_SLACK_BOT_ADMIN_CLIENT_SECRET:-caipe-platform-dev-secret}
      # Slack bot's OBO (token-exchange) credentials — separate flow, separate client.
      - KEYCLOAK_BOT_CLIENT_ID=${KEYCLOAK_BOT_CLIENT_ID:-caipe-slack-bot}
      - KEYCLOAK_BOT_CLIENT_SECRET=${KEYCLOAK_BOT_CLIENT_SECRET:-caipe-slack-bot-dev-secret}
      - SLACK_LINKING_BASE_URL=${SLACK_LINKING_BASE_URL:-http://localhost:3000}
      - SLACK_LINK_HMAC_SECRET=${SLACK_LINK_HMAC_SECRET:-${SLACK_SIGNING_SECRET:-}}
      # Spec 103: Just-In-Time Keycloak user creation. Default ON in dev so
      # first-time Slack DMs work without an admin handshake. Set to "false"
      # in production to require explicit web-UI onboarding before Slack use,
      # in which case the bot falls back to sending an HMAC-signed linking
      # URL. Reuses the KEYCLOAK_SLACK_BOT_ADMIN_* credentials above; no new
      # secret to wire (see plan.md R-8 for the accepted trade-off).
      - SLACK_JIT_CREATE_USER=${SLACK_JIT_CREATE_USER:-true}
      # Optional comma-separated allowlist of email domains eligible for JIT.
      # Empty/unset => any domain. Recommended for prod when the federated
      # IdP can return non-corporate emails.
      - SLACK_JIT_ALLOWED_EMAIL_DOMAINS=${SLACK_JIT_ALLOWED_EMAIL_DOMAINS:-}
    depends_on:
      caipe-ui:
        condition: service_started
        required: false
      caipe-mongodb:
        condition: service_started
    profiles:
      - slack-bot
      - all-integrations

  ####################################################################################################
  #                                 Webex Bot Integration                                            #
  ####################################################################################################
  webex-bot:
    build:
      context: .
      dockerfile: build/Dockerfile.webex-bot
    image: ghcr.io/cnoe-io/caipe-webex-bot:${IMAGE_TAG:-localtag}
    container_name: webex-bot
    volumes:
      - ./ai_platform_engineering/integrations/webex_bot:/app/ai_platform_engineering/integrations/webex_bot
      - /app/ai_platform_engineering/integrations/webex_bot/.venv
    command:
      - sh
      - -c
      - |
        if [ "${DEV_HOT_RELOAD:-false}" = "true" ]; then
          cd /app/ai_platform_engineering/integrations/webex_bot && uv sync && exec uv run watchfiles --filter python 'python -m ai_platform_engineering.integrations.webex_bot.main' /app/ai_platform_engineering/integrations/webex_bot
        else
          exec python -m ai_platform_engineering.integrations.webex_bot.main
        fi
    env_file: [.env]
    ports: ["8032:3002"]
    environment:
      - DEV_HOT_RELOAD=${DEV_HOT_RELOAD:-false}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - WEBEX_INTEGRATION_BOT_ACCESS_TOKEN=${WEBEX_INTEGRATION_BOT_ACCESS_TOKEN:-}
      - WEBEX_WORKSPACE_ALIAS=${WEBEX_WORKSPACE_ALIAS:-CAIPE-WEBEX}
      - WEBEX_WORKSPACE_ID=${WEBEX_WORKSPACE_ID:-}
      - CAIPE_API_URL=${CAIPE_API_URL:-http://caipe-ui:3000}
      - CAIPE_UI_BASE_URL=${CAIPE_UI_BASE_URL:-http://localhost:3000}
      - MONGODB_URI=mongodb://admin:changeme@caipe-mongodb:27017
      - MONGODB_DATABASE=caipe
      - WEBEX_AGENT_ROUTES_MODE=${WEBEX_AGENT_ROUTES_MODE:-db_prefer}
      - WEBEX_THREAD_CONTEXT_ENABLED=${WEBEX_THREAD_CONTEXT_ENABLED:-true}
      - WEBEX_THREAD_CONTEXT_MAX_MESSAGES=${WEBEX_THREAD_CONTEXT_MAX_MESSAGES:-10}
      - WEBEX_THREAD_CONTEXT_MAX_CHARS=${WEBEX_THREAD_CONTEXT_MAX_CHARS:-4000}
      - WEBEX_AUTO_ASSIGN_UNMAPPED_SPACES=${WEBEX_AUTO_ASSIGN_UNMAPPED_SPACES:-false}
      - WEBEX_DEFAULT_TEAM_SLUG=${WEBEX_DEFAULT_TEAM_SLUG:-}
      - WEBEX_DEFAULT_AGENT_ID=${WEBEX_DEFAULT_AGENT_ID:-}
      - WEBEX_ADMIN_API_ENABLED=${WEBEX_ADMIN_API_ENABLED:-false}
      - WEBEX_ADMIN_API_HOST=${WEBEX_ADMIN_API_HOST:-0.0.0.0}
      - WEBEX_ADMIN_API_PORT=${WEBEX_ADMIN_API_PORT:-3002}
      - WEBEX_ADMIN_JWT_ISSUER=${WEBEX_ADMIN_JWT_ISSUER:-http://localhost:7080/realms/caipe}
      - WEBEX_ADMIN_JWKS_URL=${WEBEX_ADMIN_JWKS_URL:-http://keycloak:7080/realms/caipe/protocol/openid-connect/certs}
      - WEBEX_ADMIN_JWT_AUDIENCE=${WEBEX_ADMIN_JWT_AUDIENCE:-caipe-webex-bot-admin}
      - WEBEX_ADMIN_ALLOWED_CLIENT_IDS=${WEBEX_ADMIN_ALLOWED_CLIENT_IDS:-caipe-ui}
      - OPENFGA_HTTP=${OPENFGA_HTTP:-http://openfga:8080}
      - OPENFGA_STORE_NAME=${OPENFGA_STORE_NAME:-caipe-openfga}
      - OPENFGA_STORE_ID=${OPENFGA_STORE_ID:-}
      - KEYCLOAK_URL=${KEYCLOAK_URL:-http://keycloak:7080}
      - KEYCLOAK_REALM=${KEYCLOAK_REALM:-caipe}
      - KEYCLOAK_WEBEX_BOT_ADMIN_CLIENT_ID=${KEYCLOAK_WEBEX_BOT_ADMIN_CLIENT_ID:-caipe-platform}
      - KEYCLOAK_WEBEX_BOT_ADMIN_CLIENT_SECRET=${KEYCLOAK_WEBEX_BOT_ADMIN_CLIENT_SECRET:-caipe-platform-dev-secret}
      - KEYCLOAK_WEBEX_BOT_CLIENT_ID=${KEYCLOAK_WEBEX_BOT_CLIENT_ID:-caipe-webex-bot}
      - KEYCLOAK_WEBEX_BOT_CLIENT_SECRET=${KEYCLOAK_WEBEX_BOT_CLIENT_SECRET:-caipe-webex-bot-dev-secret}
      - WEBEX_LINKING_BASE_URL=${WEBEX_LINKING_BASE_URL:-http://localhost:3000}
      - WEBEX_LINK_HMAC_SECRET=${WEBEX_LINK_HMAC_SECRET:-}
    depends_on:
      caipe-ui:
        condition: service_started
        required: false
      caipe-mongodb:
        condition: service_started
    profiles:
      - webex-bot
      - all-integrations

  ####################################################################################################
  #                                  Enterprise RBAC (098)                                       #
  ####################################################################################################
  # Keycloak + OpenFGA + AgentGateway for Enterprise RBAC.
  # Usage (Slack + RAG RBAC — copy deploy/rbac/.env.rbac.example into .env and set embeddings keys):
  #   COMPOSE_PROFILES=rbac,rag,slack-bot,caipe-supervisor,caipe-mongodb \
  #     docker compose -f docker-compose.dev.yaml up -d
  #
  # RAG: http://localhost:9446  (requires Milvus stack + embedding provider env, e.g. AZURE_OPENAI_*)
  # Slack bot: socket mode or http://localhost:8030 — set SLACK_* tokens in .env
  #
  # Admin console: http://localhost:7080/admin (admin / admin)
  # OpenFGA API: http://localhost:18080  Playground: http://localhost:13002
  # Agent Gateway proxy: http://localhost:4000
  # Agent Gateway admin UI: http://localhost:15000/ui
  #
  # Test personas (see deploy/keycloak/realm-config.json):
  #   admin-user / admin       → admin, chat_user
  #   standard-user / standard → chat_user, team_member
  #   kb-admin-user / kbadmin  → chat_user, team_member, kb_admin
  #   denied-user / denied     → (no roles)
  #   org-b-user / orgb        → chat_user (tenant: globex)

  # Durable Postgres for Keycloak.
  #
  # Why this exists: Keycloak's default `start-dev` uses an embedded H2 DB
  # written under /opt/keycloak/data/h2. Without an external DB AND a named
  # volume, every `docker compose up` on a recreated container wipes ALL
  # Keycloak state — users, role assignments, per-team client scopes, the
  # `team_member:<slug>` realm roles, Slack identity links (slack_user_id
  # user attribute), token-exchange grants, etc. Mongo still holds the team
  # docs, so the system silently drifts: BFF startup re-creates the
  # `team-<slug>` client scopes (via syncTeamScopesOnStartup), but realm
  # roles + role assignments + identity links are NOT reconciled, leading
  # to OBO failures like "I couldn't start your CAIPE session".
  #
  # Putting Keycloak on Postgres with a named volume makes the dev stack
  # behave like prod: `docker compose down && up` preserves state;
  # `docker compose down -v` is the only destructive op.
  keycloak-postgres:
    image: postgres:16-alpine
    container_name: keycloak-postgres
    restart: unless-stopped
    environment:
      POSTGRES_USER: keycloak
      POSTGRES_PASSWORD: keycloak
      POSTGRES_DB: keycloak
    volumes:
      - keycloak_postgres_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U keycloak -d keycloak"]
      interval: 5s
      timeout: 5s
      retries: 20
      start_period: 10s
    profiles:
      - rbac

  keycloak:
    image: quay.io/keycloak/keycloak:26.3
    container_name: keycloak
    environment:
      # Keycloak 26 requires hostname as a full URL when backchannel-dynamic
      # is enabled. Frontchannel/browser URLs use this base; backchannel URLs
      # are rewritten to use the request's hostname (so caipe-ui calling
      # http://keycloak:7080/.well-known/... gets back token_endpoint that
      # also uses keycloak:7080 — fixing ECONNREFUSED on server-to-server
      # token exchange).
      # See: https://www.keycloak.org/server/hostname#_backchannel_dynamic
      KC_HOSTNAME: "http://localhost:7080"
      KC_HOSTNAME_STRICT: "false"
      KC_HOSTNAME_BACKCHANNEL_DYNAMIC: "true"
      KEYCLOAK_ADMIN: admin
      KEYCLOAK_ADMIN_PASSWORD: admin
      KC_HEALTH_ENABLED: "true"
      KC_LOG_LEVEL: info
      # Persist KC state in Postgres instead of the ephemeral embedded H2 DB.
      # Without this, every container recreate wipes users, roles, role
      # assignments, per-team client scopes, identity links, etc. — and
      # Mongo's team docs end up referencing KC objects that no longer exist.
      KC_DB: postgres
      KC_DB_URL: jdbc:postgresql://keycloak-postgres:5432/keycloak # gitleaks:allow - local dev database URL
      KC_DB_USERNAME: keycloak
      KC_DB_PASSWORD: keycloak
      # Cap JVM heap so KC plays nicely on a busy dev laptop. Default
      # `start-dev` lets the JVM scale to 25% of host memory, which on a
      # heavily-loaded compose stack can trip Docker Desktop's OOM killer
      # during schema init (exit 137). 768m headroom is plenty for dev.
      JAVA_OPTS_KC_HEAP: "-Xms256m -Xmx768m -XX:MaxMetaspaceSize=256m"
    # `--import-realm` is still safe with a durable DB: KC only imports on
    # an empty realm, so subsequent boots become idempotent no-ops and your
    # runtime mutations (teams, role assignments, identity links) survive.
    command: ["start-dev", "--import-realm", "--http-port", "7080", "--https-port", "7443", "--features=token-exchange,admin-fine-grained-authz:v1"]
    volumes:
      - ./charts/ai-platform-engineering/charts/keycloak/realm-config.json:/opt/keycloak/data/import/realm-config.json:ro
      - ./deploy/keycloak/themes/caipe:/opt/keycloak/themes/caipe:ro
    ports:
      # Keep direct Keycloak access local to the Docker host. Public/browser
      # ingress should go through the HTTPS reverse proxy.
      - "127.0.0.1:7080:7080"
      - "127.0.0.1:7443:7443"
    depends_on:
      keycloak-postgres:
        condition: service_healthy
    # Wait until management is up AND the realm JWKS route returns 200 (TCP-only on 7080 was
    # too early — realm import can lag, leaving the fetcher in an endless curl loop).
    healthcheck:
      test:
        - CMD-SHELL
        - >-
          bash -c "exec 3<>/dev/tcp/127.0.0.1/9000 && exec 3<&- && exec 3>&- &&
          exec 3<>/dev/tcp/127.0.0.1/7080 &&
          printf $$'GET /realms/caipe/protocol/openid-connect/certs HTTP/1.0\r\nHost: localhost:7080\r\nConnection: close\r\n\r\n' >&3 &&
          head -n 1 <&3 | grep -q ' 200'"
      interval: 10s
      timeout: 10s
      retries: 25
      start_period: 120s
    profiles:
      - rbac

  # One-shot init container: creates the upstream OIDC identity-provider
  # broker in Keycloak from IDP_* env vars (set in .env).
  # Skips silently when IDP_ISSUER is unset — local-only users + test
  # personas still work via direct Keycloak login.
  keycloak-init:
    image: ${KEYCLOAK_INIT_IMAGE:-caipe/keycloak-init:dev}
    build:
      context: .
      dockerfile: build/Dockerfile.keycloak-init
    container_name: keycloak-init
    entrypoint: ["/bin/sh", "/scripts/init-idp.sh"]
    volumes:
      - ./charts/ai-platform-engineering/charts/keycloak/scripts/init-idp.sh:/scripts/init-idp.sh:ro
    environment:
      KEYCLOAK_ADMIN: admin
      KEYCLOAK_ADMIN_PASSWORD: admin
      IDP_ALIAS: ${IDP_ALIAS:-}
      IDP_DISPLAY_NAME: ${IDP_DISPLAY_NAME:-}
      IDP_ISSUER: ${IDP_ISSUER:-}
      IDP_CLIENT_ID: ${IDP_CLIENT_ID:-}
      IDP_CLIENT_SECRET: ${IDP_CLIENT_SECRET:-}
      IDP_ACCESS_GROUP: ${IDP_ACCESS_GROUP:-}
      IDP_ADMIN_GROUP: ${IDP_ADMIN_GROUP:-}
      KEYCLOAK_FORCE_IDP_REDIRECT: ${KEYCLOAK_FORCE_IDP_REDIRECT:-false}
      KEYCLOAK_ADMIN_FRONTEND_URL: ${KEYCLOAK_ADMIN_FRONTEND_URL:-}
      CAIPE_UNSAFE_RBAC_BYPASS: ${CAIPE_UNSAFE_RBAC_BYPASS:-false}
      SLACK_BOT_ADMIN_AUDIENCE: ${SLACK_BOT_ADMIN_AUDIENCE:-caipe-slack-bot-admin}
      KC_REALM: caipe
      # Spec 104: comma-separated emails that get the admin_user role and the
      # demo team/agent/tool bundle on first run of init-idp.sh. Sourced from
      # the same env var the UI uses for bootstrap admin bypass.
      BOOTSTRAP_ADMIN_EMAILS: ${BOOTSTRAP_ADMIN_EMAILS:-}
    network_mode: "service:keycloak"
    depends_on:
      keycloak:
        condition: service_healthy
    # `on-failure` so transient KC unavailability (start_period race) is
    # auto-retried instead of leaving the realm half-seeded. Both init
    # scripts are idempotent — re-running is safe. Use `make rbac-reinit`
    # to force a re-run after the container has already exited cleanly
    # (e.g. after `docker compose restart keycloak`, where the existing
    # exited init container does NOT auto-rerun).
    restart: on-failure
    profiles:
      - rbac

  # One-shot init container: configures token exchange + impersonation
  # permissions on caipe-slack-bot so the Slack bot can mint JWTs on
  # behalf of users (RFC 8693).  Idempotent — safe to re-run.
  keycloak-init-token-exchange:
    image: ${KEYCLOAK_INIT_IMAGE:-caipe/keycloak-init:dev}
    build:
      context: .
      dockerfile: build/Dockerfile.keycloak-init
    container_name: keycloak-init-token-exchange
    entrypoint: ["/bin/sh", "/scripts/init-token-exchange.sh"]
    volumes:
      - ./deploy/keycloak/init-token-exchange.sh:/scripts/init-token-exchange.sh:ro
    environment:
      KEYCLOAK_ADMIN: admin
      KEYCLOAK_ADMIN_PASSWORD: admin
      KC_REALM: caipe
      KC_BOT_CLIENT_ID: caipe-slack-bot
      KC_SSL_REQUIRED: none
      # Seed the Slack bot's service-account read grant on
      # system_config:platform_settings so it can read platform-wide settings
      # (default agent, VictorOps agent) from the BFF. The BFF graphs the bot's
      # client-credentials token as service_account:<sub>; without this grant
      # the read 403s and the bot falls back to env/YAML defaults.
      OPENFGA_HTTP: http://openfga:8080
      OPENFGA_STORE_NAME: caipe-openfga
    network_mode: "service:keycloak"
    depends_on:
      keycloak:
        condition: service_healthy
      # Run AFTER init-idp.sh so we're configuring token exchange against a
      # realm that already has the slack-bot client wired up by the IdP seed.
      keycloak-init:
        condition: service_completed_successfully
      # Need the OpenFGA store + model seeded before we can write the bot's
      # platform-settings grant tuple.
      openfga-init:
        condition: service_completed_successfully
    restart: on-failure
    profiles:
      - rbac

  # OpenFGA remote PDP for AgentGateway ext_authz.
  #
  # AgentGateway calls openfga-authz-bridge on every MCP request; OpenFGA is
  # the authorization source for the gateway path. We do not run a CEL policy
  # config bridge for AgentGateway.
  openfga-postgres:
    image: postgres:17-alpine
    container_name: openfga-postgres
    restart: unless-stopped
    profiles:
      - rbac
    environment:
      POSTGRES_USER: ${OPENFGA_POSTGRES_USER:-openfga}
      POSTGRES_PASSWORD: ${OPENFGA_POSTGRES_PASSWORD:-openfga}
      POSTGRES_DB: ${OPENFGA_POSTGRES_DB:-openfga}
    volumes:
      - openfga_postgres_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U $${POSTGRES_USER} -d $${POSTGRES_DB}"]
      interval: 5s
      timeout: 5s
      retries: 20
      start_period: 10s

  openfga-migrate:
    image: openfga/openfga:latest
    container_name: openfga-migrate
    profiles:
      - rbac
    depends_on:
      openfga-postgres:
        condition: service_healthy
    environment:
      OPENFGA_DATASTORE_ENGINE: postgres
      OPENFGA_DATASTORE_URI: postgres://${OPENFGA_POSTGRES_USER:-openfga}:${OPENFGA_POSTGRES_PASSWORD:-openfga}@openfga-postgres:5432/${OPENFGA_POSTGRES_DB:-openfga}?sslmode=disable
    command: migrate

  openfga:
    image: openfga/openfga:latest
    container_name: openfga
    restart: unless-stopped
    profiles:
      - rbac
    depends_on:
      openfga-migrate:
        condition: service_completed_successfully
    environment:
      OPENFGA_DATASTORE_ENGINE: postgres
      OPENFGA_DATASTORE_URI: postgres://${OPENFGA_POSTGRES_USER:-openfga}:${OPENFGA_POSTGRES_PASSWORD:-openfga}@openfga-postgres:5432/${OPENFGA_POSTGRES_DB:-openfga}?sslmode=disable
      OPENFGA_DATASTORE_MAX_OPEN_CONNS: "50"
      OPENFGA_PLAYGROUND_ENABLED: "true"
      OPENFGA_TRACE_ENABLED: "false"
    command: run
    ports:
      - "127.0.0.1:${OPENFGA_HTTP_PORT:-18080}:8080"
      - "127.0.0.1:${OPENFGA_GRPC_PORT:-18081}:8081"
      - "127.0.0.1:${OPENFGA_PLAYGROUND_PORT:-13002}:3000"
    healthcheck:
      test:
        - CMD
        - /usr/local/bin/grpc_health_probe
        - -addr=localhost:8081
      interval: 5s
      timeout: 30s
      retries: 10

  openfga-init:
    build:
      context: ./deploy/openfga/init
    container_name: openfga-init
    profiles:
      - rbac
    volumes:
      - ./charts/ai-platform-engineering/charts/openfga/authorization-model.json:/app/authorization-model.json:ro
    environment:
      OPENFGA_HTTP: http://openfga:8080
      OPENFGA_STORE_NAME: caipe-openfga
      # Seed a dev allow tuple with a Keycloak user `sub`. If unset, OpenFGA
      # starts deny-by-default and you can write tuples later.
      OPENFGA_SEED_SUB: ${OPENFGA_SEED_SUB:-}
    depends_on:
      openfga:
        condition: service_healthy

  openfga-authz-bridge:
    build:
      context: ./deploy/openfga/bridge
    container_name: openfga-authz-bridge
    restart: unless-stopped
    profiles:
      - rbac
    environment:
      OPENFGA_HTTP: http://openfga:8080
      OPENFGA_STORE_NAME: caipe-openfga
      OPENFGA_RELATION: can_call
      OPENFGA_OBJECT: mcp_gateway:list
      OPENFGA_BYPASS_SUBS: ${OPENFGA_BYPASS_SUBS:-}
      CAIPE_AGENT_CONTEXT_HMAC_SECRET: ${CAIPE_AGENT_CONTEXT_HMAC_SECRET:-}
      MONGODB_URI: ${MONGODB_URI:-mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@caipe-mongodb:27017/${MONGODB_DATABASE:-caipe}?authSource=admin}
      MONGODB_DATABASE: ${MONGODB_DATABASE:-caipe}
      TENANT_ID: ${TENANT_ID:-default}
      AUDIT_SUBJECT_SALT: ${AUDIT_SUBJECT_SALT:-caipe-098-audit}
    ports:
      - "127.0.0.1:${OPENFGA_BRIDGE_PORT:-9100}:9100"
    healthcheck:
      test:
        - CMD
        - python
        - -c
        - "import socket; socket.create_connection(('localhost', 9100), timeout=2).close()"
      interval: 5s
      timeout: 3s
      retries: 10
    depends_on:
      openfga-init:
        condition: service_completed_successfully
      openfga:
        condition: service_healthy

  agentgateway-config-bridge:
    build:
      context: ./deploy/agentgateway
      dockerfile: Dockerfile.config-bridge
    image: ghcr.io/cnoe-io/caipe-agentgateway-config-bridge:${IMAGE_TAG:-localtag}
    container_name: agentgateway-config-bridge
    restart: unless-stopped
    read_only: true
    security_opt:
      - no-new-privileges:true
    profiles:
      - rbac
    environment:
      MONGODB_URI: ${MONGODB_URI:-mongodb://${MONGODB_ROOT_USERNAME:-admin}:${MONGODB_ROOT_PASSWORD:-changeme}@caipe-mongodb:27017/${MONGODB_DATABASE:-caipe}?authSource=admin}
      MONGODB_DATABASE: ${MONGODB_DATABASE:-caipe}
      AGENTGATEWAY_CONFIG_PATH: /generated/config.yaml
      AGENTGATEWAY_BOOTSTRAP_CONFIG_PATH: /bootstrap/config.yaml
      AGENTGATEWAY_ADMIN_CONFIG_URL: http://agentgateway:15000/config
      AGENTGATEWAY_CONFIG_BRIDGE_POLL_SECONDS: ${AGENTGATEWAY_CONFIG_BRIDGE_POLL_SECONDS:-5}
      LOG_LEVEL: ${AGENTGATEWAY_CONFIG_BRIDGE_LOG_LEVEL:-INFO}
    volumes:
      - agentgateway_config:/generated
      - ./deploy/agentgateway/config.yaml:/bootstrap/config.yaml:ro
    tmpfs:
      - /tmp
    healthcheck:
      test:
        - CMD
        - python
        - -c
        - "from pathlib import Path; import sys; sys.exit(0 if Path('/generated/config.yaml').is_file() and Path('/generated/config.yaml').stat().st_size > 0 else 1)"
      interval: 5s
      timeout: 3s
      retries: 12
      start_period: 5s
    depends_on:
      caipe-mongodb:
        condition: service_healthy

  agentgateway:
    image: ghcr.io/agentgateway/agentgateway:latest
    container_name: agentgateway
    restart: unless-stopped
    command: ["--file", "/etc/agentgateway/config.yaml"]
    environment:
      GITHUB_PERSONAL_ACCESS_TOKEN: ${GITHUB_PERSONAL_ACCESS_TOKEN:-}
      GITLAB_PERSONAL_ACCESS_TOKEN: ${GITLAB_PERSONAL_ACCESS_TOKEN:-${GITLAB_TOKEN:-}}
    read_only: true
    security_opt:
      - no-new-privileges:true
    profiles:
      - rbac
    depends_on:
      keycloak:
        condition: service_healthy
      # Wait for the IdP broker + spec-104 role bundle to be seeded before
      # AG starts validating tokens — otherwise the very first request after
      # `compose up` can race the seed and 403 a user who SHOULD have roles.
      keycloak-init:
        condition: service_completed_successfully
      # Token-exchange grants must exist before the Slack bot's first OBO
      # call, but AG itself doesn't need it. Listed here so that bringing
      # up AG implicitly brings up the full KC initialization chain — saves
      # operators from having to remember `--profile rbac` orders things
      # correctly. Marked optional via `restart: on-failure` on the init
      # container so a failure here doesn't block AG from coming up if the
      # token-exchange config is intentionally absent in dev.
      keycloak-init-token-exchange:
        condition: service_completed_successfully
      openfga-authz-bridge:
        condition: service_healthy
      agentgateway-config-bridge:
        condition: service_healthy
    ports:
      - "4000:4000"
      - "127.0.0.1:15000:15000" # Admin UI (config.config.adminAddr)
    volumes:
      - agentgateway_config:/etc/agentgateway:ro
    tmpfs:
      - /tmp
    healthcheck:
      disable: true

  ####################################################################################################
  #                       LangGraph Persistence Backends                                         #
  ####################################################################################################
  # Redis Stack, PostgreSQL, and MongoDB for testing LangGraph checkpointer + store persistence.
  #
  # Usage (pick ONE backend):
  #
  #   Redis Stack:
  #     LANGGRAPH_CHECKPOINT_TYPE=redis \
  #     LANGGRAPH_CHECKPOINT_REDIS_URL=redis://langgraph-redis:6379 \
  #     LANGGRAPH_STORE_TYPE=redis \
  #     LANGGRAPH_STORE_REDIS_URL=redis://langgraph-redis:6379 \
  #     docker compose -f docker-compose.dev.yaml --profile langgraph-redis up
  #
  #   PostgreSQL:
  #     LANGGRAPH_CHECKPOINT_TYPE=postgres \
  #     LANGGRAPH_CHECKPOINT_POSTGRES_DSN=postgresql://langgraph:langgraph@langgraph-postgres:5432/langgraph \
  #     LANGGRAPH_STORE_TYPE=postgres \
  #     LANGGRAPH_STORE_POSTGRES_DSN=postgresql://langgraph:langgraph@langgraph-postgres:5432/langgraph \
  #     docker compose -f docker-compose.dev.yaml --profile langgraph-postgres up
  #
  #   MongoDB:
  #     LANGGRAPH_CHECKPOINT_TYPE=mongodb \
  #     LANGGRAPH_CHECKPOINT_MONGODB_URI=mongodb://langgraph-mongodb:27017 \
  #     LANGGRAPH_STORE_TYPE=mongodb \
  #     LANGGRAPH_STORE_MONGODB_URI=mongodb://langgraph-mongodb:27017 \
  #     docker compose -f docker-compose.dev.yaml --profile langgraph-mongodb up
  #
  #   Mixed (MongoDB checkpointer + Redis store, sharing UI MongoDB):
  #     LANGGRAPH_CHECKPOINT_TYPE=mongodb \
  #     LANGGRAPH_CHECKPOINT_MONGODB_URI="mongodb://admin:changeme@caipe-mongodb:27017/caipe?authSource=admin" \
  #     LANGGRAPH_CHECKPOINT_MONGODB_DB_NAME=caipe \
  #     LANGGRAPH_CHECKPOINT_MONGODB_COLLECTION=checkpoints_conversation \
  #     LANGGRAPH_CHECKPOINT_MONGODB_WRITES_COLLECTION=checkpoint_writes_conversation \
  #     LANGGRAPH_STORE_TYPE=redis \
  #     LANGGRAPH_STORE_REDIS_URL=redis://langgraph-redis:6379 \
  #     docker compose -f docker-compose.dev.yaml --profile deps --profile langgraph-redis up

  langgraph-redis:
    image: redis:8.0-alpine
    container_name: langgraph-redis
    ports:
      - "6381:6379"
    volumes:
      - langgraph_redis_data:/data
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 10s
      timeout: 5s
      retries: 5
    restart: unless-stopped
    profiles:
      - langgraph-redis

  langgraph-postgres:
    image: postgres:16-alpine
    container_name: langgraph-postgres
    ports:
      - "5433:5432"
    environment:
      - POSTGRES_USER=langgraph
      - POSTGRES_PASSWORD=langgraph
      - POSTGRES_DB=langgraph
    volumes:
      - langgraph_postgres_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U langgraph"]
      interval: 10s
      timeout: 5s
      retries: 5
    restart: unless-stopped
    profiles:
      - langgraph-postgres

  langgraph-mongodb:
    image: mongo:7
    container_name: langgraph-mongodb
    ports:
      - "27018:27017"
    volumes:
      - langgraph_mongodb_data:/data/db
    healthcheck:
      test: ["CMD", "mongosh", "--eval", "db.adminCommand('ping')"]
      interval: 10s
      timeout: 5s
      retries: 5
    restart: unless-stopped
    profiles:
      - langgraph-mongodb

  ####################################################################################################
  #                                      Skill Scanner                                               #
  ####################################################################################################
  # cisco-ai-defense/skill-scanner running its built-in REST API server. The
  # CAIPE UI POSTs zipped SKILL packages to /scan-upload to get static +
  # (optionally) LLM-driven safety findings.
  #
  # Why standalone (not co-located with caipe-supervisor):
  #   - Avoid coupling MAS lifecycle to scanner pip installs.
  #   - Independent upgrade path (scanner releases ship frequently).
  #   - Stays unauthenticated only on the internal docker network — never
  #     map this port on a public interface.
  skill-scanner:
    build:
      context: .
      dockerfile: build/Dockerfile.skill-scanner
      args:
        # Bumping requires verifying /scan-upload response shape vs
        # ui/src/lib/skill-scan.ts.
        - SKILL_SCANNER_VERSION=${SKILL_SCANNER_VERSION:-2.0.11}
    image: ghcr.io/cnoe-io/skill-scanner:${IMAGE_TAG:-localtag}
    container_name: skill-scanner
    # Bind to 127.0.0.1 only — the API has no auth and accepts arbitrary
    # ZIPs, so it must never be reachable from outside the host.
    ports:
      - "127.0.0.1:8765:8000"
    environment:
      # Optional LLM-analyzer credentials. Leave unset to run static-only
      # scans (still useful for catching obvious unsafe patterns).
      - SKILL_SCANNER_LLM_API_KEY=${SKILL_SCANNER_LLM_API_KEY:-}
      - SKILL_SCANNER_LLM_MODEL=${SKILL_SCANNER_LLM_MODEL:-}
    healthcheck:
      test: ["CMD-SHELL", "python -c \"import urllib.request,sys;sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:8000/health',timeout=2).status==200 else 1)\""]
      # Long interval (5m) is intentional: skill-scanner-api is a thin Uvicorn
      # process with no background workers — once `start_period` passes it
      # effectively never goes unhealthy on its own. The 15s default was
      # spamming `INFO ... "GET /health HTTP/1.1" 200 OK` into `docker compose`
      # logs every interval. 5m is enough to catch a crashed/looping pod for
      # local dev without drowning the console; production uses the Helm chart
      # which sets its own probe cadence.
      interval: 5m
      timeout: 5s
      retries: 3
      start_period: 10s
    restart: unless-stopped
    profiles:
      - skill-scanner
      - caipe-ui

volumes:
  mongodb_data:
    driver: local
  mongodb_config:
    driver: local
  langfuse_postgres_data:
    driver: local
  langfuse_clickhouse_data:
    driver: local
  langfuse_clickhouse_logs:
    driver: local
  langfuse_minio_data:
    driver: local
  langfuse_redis_data:
    driver: local
  rag_redis_data:
    driver: local
  milvus_etcd:
    driver: local
  milvus_minio:
    driver: local
  milvus_data:
    driver: local
  langgraph_redis_data:
    driver: local
  langgraph_postgres_data:
    driver: local
  langgraph_mongodb_data:
    driver: local
  openfga_postgres_data:
    driver: local
  agentgateway_config:
    driver: local
  # Durable Keycloak state (users, role assignments, per-team client scopes,
  # `team_member:<slug>` realm roles, identity links, token-exchange grants).
  # See the keycloak-postgres service for the rationale.
  keycloak_postgres_data:
    driver: local