Skip to content

OpenClaw

Lightweight bridge server that orchestrates AI coding agents on Kubernetes. Provides a stateless, Git-native workflow where agents clone repositories, make changes, commit, and push within ephemeral containers.
Category ai-agent
Official Docs OpenClaw Repository
CLI Install ./cli ai-agent openclaw install
CLI Uninstall ./cli ai-agent openclaw uninstall
Namespace openclaw

Overview

OpenClaw deploys a central bridge server that: - Task Routing: Receives task requests and routes to appropriate agent containers - Real-Time Streaming: Streams agent responses back to clients - LiteLLM Integration: Routes all LLM calls through LiteLLM gateway - Langfuse Observability: Sends traces for monitoring and debugging - Open WebUI Integration: Works as a backend for Open WebUI pipe functions - Stateless Design: Git is the durable store; containers are ephemeral

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         Open WebUI                               │
│                    (GUI, Pipe Functions)                         │
└────────────────────────────┬────────────────────────────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────────┐
│                    OpenClaw Bridge Server                        │
│                  (Task Router, Orchestrator)                     │
│                  http://openclaw.openclaw:8080                   │
└─────────┬───────────────────────────────────────────┬───────────┘
          │                                           │
          ▼                                           ▼
┌──────────────────────┐                  ┌──────────────────────┐
│  Doc Writer Agent    │                  │   DevOps Agent       │
│  (Example)           │                  │   (Example)          │
└──────────────────────┘                  └──────────────────────┘
          │                                           │
          └───────────────────┬───────────────────────┘
                              ▼
                    ┌──────────────────────┐
                    │   LiteLLM Gateway    │
                    └──────────────────────┘

Installation

Prerequisites

  • LiteLLM installed (./cli ai-gateway litellm install)
  • LITELLM_API_KEY set in .env

Install

./cli ai-agent openclaw install

The installer: 1. Deploys the bridge server using a pre-built public ECR image 2. Creates Kubernetes Service at http://openclaw.openclaw:8080 3. Auto-detects and configures Langfuse integration if available

Verification

# Check pods
kubectl get pods -n openclaw

# Check service
kubectl get svc -n openclaw

# Check logs
kubectl logs -n openclaw -l app=openclaw

# Test health endpoint
kubectl port-forward -n openclaw svc/openclaw 8080:8080
curl http://localhost:8080/health

Expected output:

{"status":"ok"}

Configuration

config.json

{
  "ai-agent": {
    "openclaw": {
      "env": {
        "LITELLM_MODEL_NAME": "bedrock/claude-4.5-sonnet",
        "OPENCLAW_GATEWAY_TOKEN": "openclaw-gateway-token"
      }
    }
  }
}

Environment Variables

Variable Description Default
OPENCLAW_GATEWAY_TOKEN Authentication token openclaw-gateway-token
LITELLM_BASE_URL LiteLLM API endpoint http://litellm.litellm:4000
LITELLM_API_KEY LiteLLM API key From .env
LITELLM_MODEL_NAME Default model bedrock/claude-4.5-sonnet
LANGFUSE_HOST Langfuse endpoint (optional) http://langfuse-web.langfuse:3000
LANGFUSE_PUBLIC_KEY Langfuse public key (optional) From .env
LANGFUSE_SECRET_KEY Langfuse secret key (optional) From .env

Switching Models

Update LITELLM_MODEL_NAME in config.json and reinstall:

./cli ai-agent openclaw uninstall
./cli ai-agent openclaw install

Integration

Langfuse

If Langfuse is installed, OpenClaw automatically sends traces for: - Task submissions and agent responses - LLM API calls with token usage - Error events and latency metrics

Open WebUI

Open WebUI connects to OpenClaw via pipe functions to provide agent capabilities in the chat interface.

Examples

See the following examples for complete agent implementations:

Cost Optimization

  • Stateless containers: No persistent storage (Git is the durable store)
  • Spot instances: Karpenter provisions Spot ARM64 nodes (up to 90% savings)
  • Ephemeral execution: Jobs terminate after completion
  • Estimated cost: ~$0 at rest, ~$0.02-0.05 per task (excluding LLM API costs)

Security

  • Authentication: Optional Bearer token auth via OPENCLAW_GATEWAY_TOKEN
  • Network: Bridge server only accessible within cluster (ClusterIP service)
  • RBAC: ServiceAccount with minimal permissions

Troubleshooting

Bridge server not starting

# Check pod events
kubectl describe pod -n openclaw -l app=openclaw

# Check logs
kubectl logs -n openclaw -l app=openclaw

Cannot connect to LiteLLM

# Check LiteLLM is running
kubectl get pods -n litellm

# Test from OpenClaw pod
kubectl exec -it -n openclaw <openclaw-pod> -- \
  curl http://litellm.litellm:4000/health

Agent tasks failing

# Check agent logs
kubectl logs -n openclaw -l app=openclaw --tail=100

# Check Langfuse traces for error details

Learn More