Skip to main content

​
Overview

This template provides a production-ready OpenClaw AI assistant gateway as a Monk runnable. You can:
  • Run it directly to get a managed AI gateway for messaging platforms
  • Inherit it in your own applications to add AI assistant capabilities
  • Combine with Ollama for local LLM inference
OpenClaw is an AI assistant gateway that connects to messaging platforms (WhatsApp, Telegram, Discord, Slack, Signal) and provides AI-powered responses using various LLM providers.

​
What this template manages

  • OpenClaw gateway container (alpine/openclaw image)
  • HTTP API service on port 18789
  • Bridge service on port 18790
  • Configuration and workspace persistence
  • Health checks (readiness and liveness)
  • Optional Ollama integration for local LLMs

​
Quick start (run directly)

  1. Load templates
monk load MANIFEST
  1. Add required secrets
monk secrets add -g openclaw-gateway-token="$(openssl rand -hex 32)"
  1. Run OpenClaw gateway
monk run openclaw/gateway
Once started, the gateway is available at localhost:18789.

​
With Ollama integration

For local LLM inference without external API dependencies:
  1. Load both templates
monk load MANIFEST
  1. Add secrets
monk secrets add -g openclaw-gateway-token="$(openssl rand -hex 32)"
  1. Run the combined stack
# CPU version
monk run openclaw/stack-ollama

# GPU version (requires NVIDIA GPU)
monk run openclaw/stack-ollama-gpu
  1. Pull a model
# Fast model for CPU
monk do openclaw/ollama/pull-model model=qwen2.5:0.5b

# Or larger model for GPU
monk do openclaw/ollama/pull-model model=llama3.3

​
Configuration

Key variables you can customize: 
variables:
  openclaw-image: "alpine/openclaw:2026.2.1"  # container image
  gateway-port: 18789                          # HTTP API port
  bridge-port: 18790                           # bridge port
  gateway-bind: "lan"                          # bind mode (lan, loopback, public)
Required secrets: 
# Gateway authentication token (required)
monk secrets add -g openclaw-gateway-token="$(openssl rand -hex 32)"

# Optional: Claude session keys
monk secrets add -g claude-ai-session-key="your-key"

​
Changing the Default Model

The template comes with a pre-configured default model (qwen2.5:0.5b for CPU, llama3.3 for GPU). If you want to use a different Ollama model, you need to:
  1. Pull the new model:
monk do openclaw/ollama/pull-model model=your-model-name
  1. Update the gateway configuration by modifying openclaw.yaml:
Find the gateway-ollama section and update the models list and primary model: 
# In the files.ollama-config.contents JSON:
"models": [
  { "id": "your-model-name", "name": "Your Model Name" },
  # ... other models
]

# And update the default:
"agents": {
  "defaults": {
    "model": {
      "primary": "ollama/your-model-name"
    }
  }
}
  1. Reload and restart:
monk load MANIFEST
monk stop openclaw/stack-ollama
monk run openclaw/stack-ollama

​
Example: Using Mistral

# Pull the model
monk do openclaw/ollama/pull-model model=mistral

# Edit openclaw.yaml to add mistral to models list and set as primary:
# "models": [{ "id": "mistral", "name": "Mistral 7B" }, ...]
# "primary": "ollama/mistral"

# Reload
monk load MANIFEST
monk stop openclaw/stack-ollama
monk run openclaw/stack-ollama
Note: The model ID in the configuration must exactly match the model name in Ollama (as shown by monk do openclaw/ollama/list-models).

​
Use by inheritance

Inherit the OpenClaw runnable in your application: 
namespace: myapp

ai-gateway:
  defines: runnable
  inherits: openclaw/gateway
  variables:
    gateway-port:
      value: 8080

backend:
  defines: runnable
  containers:
    api:
      image: myorg/backend
  connections:
    openclaw:
      runnable: ai-gateway
      service: http
  variables:
    openclaw-host:
      value: <- connection-hostname("openclaw")
    openclaw-port:
      value: "18789"
Then run your application: 
monk run myapp/backend

​
Ports and connectivity

  • Service http: TCP port 18789 (Gateway API)
  • Service bridge: TCP port 18790 (Bridge connections)
  • Health endpoint: GET /health
From other runnables, use connection-hostname("\<connection-name>") to resolve the OpenClaw host.

​
Persistence

  • Configuration: ${monk-volume-path}/openclaw/config:/home/node/.openclaw
  • Workspace: ${monk-volume-path}/openclaw/workspace:/home/node/.openclaw/workspace
  • Ollama models: ${monk-volume-path}/ollama:/root/.ollama

​
Features

  • Multi-platform messaging support (WhatsApp, Telegram, Discord, Slack, Signal)
  • Multiple LLM provider support (OpenAI, Anthropic, Ollama, and more)
  • Local LLM inference with Ollama integration
  • Persistent configuration and workspaces
  • Health monitoring with readiness/liveness checks
  • CLI tools for channel management

​
Available Actions

ActionDescriptionCommand
onboardRun onboarding wizardmonk do openclaw/cli/onboard
channels-loginLogin to WhatsAppmonk do openclaw/cli/channels-login
channels-statusCheck channels statusmonk do openclaw/cli/channels-status
healthGateway health checkmonk do openclaw/cli/health
pull-modelPull Ollama modelmonk do openclaw/ollama/pull-model model=llama3.3
list-modelsList Ollama modelsmonk do openclaw/ollama/list-models
For CPU (no GPU):
ModelSizeSpeedUse Case
qwen2.5:0.5b398MB⚡⚡⚡Testing, fast responses
tinyllama637MB⚡⚡⚡Basic tasks
llama3.2:1b1.3GB⚡⚡General purpose
gemma2:2b1.6GB⚡⚡Quality responses
For GPU:
ModelSizeSpeedUse Case
llama3.34.7GB⚡⚡⚡High quality
qwen2.5-coder:32b19GB⚡⚡Code generation
deepseek-r1:32b19GB⚡⚡Reasoning tasks
  • Combine with ollama/ollama for local LLM inference
  • Use with database templates for persistent storage
  • Integrate with monitoring templates for observability

​
Troubleshooting

  • Check gateway logs:
monk logs -f openclaw/gateway
  • Check Ollama logs:
monk logs -f openclaw/ollama
  • Verify gateway health:
curl http://localhost:18789/health
  • Enter container shell:
monk shell openclaw/gateway
  • For Ollama 404 errors, ensure a model is pulled:
monk do openclaw/ollama/pull-model model=qwen2.5:0.5b