Daemon Docker Installation

What is the Agent Daemon?

Miri can use AI for agentic Taks, but AI also needs a local Connection for relevant business data, like a SAP ERP connection. For this purpose, the AI task will not be processed centrally on the Miri web platform but decentralized within the Agent Daemon. The Agent Daemon can be installed anywhere, but it should run where the system connections are possible. Miri needs no open port or inbound channel to connect to the agent. The agent connects via outbound to the central Miri application.

The daemon is available as a Docker image to limit the capabilities of AI and has an additional native security sandbox within the AI agent, too. This is the recommended approach for productive AI Workloads within an enterprise.

Prerequisites

Docker and Docker Compose are installed on the host
Network access to the Agent Orchestrator backend
A registration token from the Agent Orchestrator UI
An API Key from an AI Model Provider

System Ressources

The Agents can run on any local devices like a laptop but for stable access, it is recommended to install it on a dedicated host with

CPU Requirements: 2 VCPUs
Memory (RAM): 4 GB
Recommended OS: Linux (Debian, Ubuntu, etc)
Network Ingres: No requirements, No Public IP or Proxy needed
Network Outgres: Internet Access required – Communication over HTTPS 443 to miri.simplifier.cloud

Step 1: Prepare Your API Key from your AI Model Provider

Currently we support

Antrophic (via API Key)
OpenRouter
Github Enterprise

Step 2 Getting a Registration Token

Login to Miri Webinterface as Administrator
Navigate to Runtimes in the sidebar.
Press the Registration Token button.
Copy the token shown in the dialog.

If no runtime exists yet, create one first via the New Runtime button on the Agent Management page.

If you don’t have admin rights, request a token from your Simplifier Contact Partner.

Step 3: Prepare the Compose File

Create an environment file and a Docker Compose configuration:

# Create the environment file
cat > .env <<'EOF'
AGENTHUB_SERVER_URL=https://miri.simplifier.cloud
AGENTHUB_TOKEN=<Enter your Registration Token here>
AGENTHUB_RUNTIME_NAME=Miri Agent Runtime
AGENTHUB_MAX_CONCURRENT=4
ANTHROPIC_API_KEY=
OPENROUTER_API_KEY=
GITHUB_ENTERPRISE_AI_KEY=
EOF

# Create docker-compose.yml
cat > docker-compose.yml <<'EOF'
services:
  daemon:
    image: simplifierag/miri-daemon:latest
    container_name: miri-daemon
    environment:
      AGENTHUB_SERVER_URL: "${AGENTHUB_SERVER_URL}"
      AGENTHUB_TOKEN: "${AGENTHUB_TOKEN}"
      AGENTHUB_WORKSPACE_ID: "${AGENTHUB_WORKSPACE_ID:-}"
      AGENTHUB_RUNTIME_NAME: "${AGENTHUB_RUNTIME_NAME:-Docker Runtime}"
      AGENTHUB_MAX_CONCURRENT: "${AGENTHUB_MAX_CONCURRENT:-4}"
      AGENTHUB_POLL_INTERVAL: "${AGENTHUB_POLL_INTERVAL:-3s}"
      AGENTHUB_IDLE_WATCHDOG: "${AGENTHUB_IDLE_WATCHDOG:-}"
      # Provider API keys for opencode — the daemon spawns opencode with `env = dict(os.environ)`,
      # so these are passed straight through to every agent subprocess. 
      # With a key set, agent can use that provider's models (set AGENTHUB_OPENCODE_MODEL to e.g.
      # "anthropic/claude-sonnet-4-5" or "openrouter/anthropic/claude-sonnet-4-5", or use per-agent
      # model overrides). Multiple keys can be set simultaneously — agent discovers all available
      # providers' models on startup.
      ANTHROPIC_API_KEY: "${ANTHROPIC_API_KEY:-}"
      OPENROUTER_API_KEY: "${OPENROUTER_API_KEY:-}"
      GITHUB_ENTERPRISE_AI_KEY: "${GITHUB_ENTERPRISE_AI_KEY:-}"
    volumes:
      - daemon-workspaces:/root/agenthub_workspaces
    labels:
      - wud.watch.digest=true
    restart: unless-stopped
  # WUD — What's Up Docker? Watches container images for new versions and
  # exposes a web UI for status. Checks every 6 hours.
  wud:
    image: getwud/wud:latest
    container_name: miri-wud
    environment:
      WUD_WATCHER_LOCAL_WATCHBYDEFAULT: "true"
      WUD_WATCHER_LOCAL_CRON: "0 */6 * * *"
      WUD_SERVER_ENABLED: "false"
    volumes:
      - "/var/run/docker.sock:/var/run/docker.sock:ro"
    restart: unless-stopped 
volumes:
  daemon-workspaces:
EOF

After the config is completed, start the Daemon


# Start the daemon
docker compose up -d

Docker Environment Variables

Variable	Default	Description
`AGENTHUB_SERVER_URL`	(required)	Backend base URL
`AGENTHUB_TOKEN`	(required)	Runtime registration token
`AGENTHUB_WORKSPACE_ID`	(none)	Pin to a specific workspace
`AGENTHUB_RUNTIME_NAME`	`Docker Runtime`	Display name in registry
`AGENTHUB_MAX_CONCURRENT`	`4`	Maximum parallel runs
`AGENTHUB_POLL_INTERVAL`	`3s`	Seconds between work-claim polls
`AGENTHUB_IDLE_WATCHDOG`	(disabled)	Auto-restart after idle period
`ANTHROPIC_API_KEY`	(disabled)	Antropic API Key for accessing Models like Sonnet, Opus etc
`OPENROUTER_API_KEY`	(disabled)	OpenRouter API Key for accessing over 500 AI Models
`GITHUB_ENTERPRISE_AI_KEY`	(disabled)	API Key for GitHub Enterprise Subscription

Managing the Agent Daemon

# View logs
docker compose logs -f

# Restart
docker compose restart

# Stop
docker compose down

# Update to latest image
docker compose pull
docker compose up -d

Multiple Daemons

To run multiple daemons on the same Docker host, use different project names and container names:

# Daemon 1
docker compose -p daemon1 up -d

# Daemon 2 (with a different runtime name)
Edit the env. file

AGENTHUB_RUNTIME_NAME="GPU Worker"

Restart the Container

docker compose -p daemon2 up -d

Verifying Installation

After starting the daemon, check the Runtimes page in Agent Orchestrator. Your new runtime should appear with a green Connected status badge within a few seconds. Recent runs will show up in the Activity column.

Troubleshooting

Symptom	Check
Daemon won’t start	`journalctl -u agenthub-daemon -n 50`
No runs claimed	Verify the workspace has queued tasks with agent assignments
Docker: container exits immediately	Check `docker compose logs daemon` for registration errors
Docker: 403 registration error	Verify `AGENTHUB_TOKEN` is a valid, non-expired registration token