OpenClaw Integration

Monitor your OpenClaw personal AI assistant with AgentGazer — one-click setup, full control

Overview

OpenClaw is an open-source personal AI assistant. By routing OpenClaw's LLM requests through AgentGazer, you get:

• Cost tracking — See exactly how much each conversation costs
• Latency monitoring — Detect performance issues
• Error alerts — Know immediately when API calls fail
• Budget controls — Set daily spending limits

Quick Start (Dashboard)

The easiest way to integrate is through the AgentGazer Dashboard.

Step 1: Start AgentGazer

agentgazer start

Step 2: Configure Providers

1. Open Dashboard at http://localhost:18880
2. Go to Providers page
3. Add your LLM provider API keys (Anthropic, OpenAI, etc.)

Step 3: OpenClaw Integration Page

1. Go to OpenClaw page in the sidebar
2. Verify your providers are listed under "Prerequisites"
3. Set Proxy Host (default: localhost:18900, use internal IP for network access)
4. Enter an Agent Name (e.g., openclaw)
5. Click Apply Configuration

This automatically writes to ~/.openclaw/openclaw.json.

Step 4: Restart OpenClaw

openclaw restart

Step 5: Send a Test Message

Send a test message through OpenClaw (Discord, Telegram, etc.), then check the Agents page — your OpenClaw agent should appear.

Step 6: Configure Model Routing

1. Go to Agents → openclaw → Model Settings
2. For the agentgazer provider, configure:
   • Model Override: The actual model to use (e.g., claude-sonnet-4-20250514)
   • Target Provider: The actual provider (e.g., anthropic)

How It Works

┌─────────────────────────────────────────────────────────┐
│                    Your Machine                          │
│                                                          │
│  ┌─────────────┐     ┌──────────────────┐               │
│  │  OpenClaw   │────▶│  AgentGazer      │               │
│  │             │     │  Proxy :18900    │               │
│  │  baseUrl →  │     │                  │               │
│  │  :18900     │     │  Auto-captures:  │               │
│  └─────────────┘     │  - tokens        │               │
│                      │  - cost          │               │
│                      │  - latency       │               │
│                      └────────┬─────────┘               │
│                               │                         │
│                   ┌───────────▼───────────┐             │
│                   │  LLM Provider APIs    │             │
│                   │  (Anthropic, OpenAI)  │             │
│                   └───────────────────────┘             │
└─────────────────────────────────────────────────────────┘

The proxy intercepts requests, extracts metrics, and forwards to the real provider. Prompt content is never stored — only token counts, latency, and cost.

Setting Up Alerts

After integration, configure alerts for your OpenClaw agent:

1. Go to Alerts page
2. Click New Alert Rule
3. Select agent: openclaw

Recommended Alert Rules

Type	Settings	Purpose
Agent Down	10 minutes	Detect when OpenClaw stops running
Error Rate	15% / 10 min	Detect API failures
Budget	$20/day	Prevent runaway costs

Governance Features

Control OpenClaw's LLM usage from the Dashboard:

Feature	Description
Active Toggle	Disable agent to block all requests
Budget Limit	Set daily spending cap
Allowed Hours	Restrict when LLM calls are allowed
Kill Switch	Auto-disable on detected infinite loops
Model Override	Force cheaper models

Configure these in Agents → openclaw → Policy Settings.

Troubleshooting

Problem	Solution
OpenClaw calls not appearing	Check baseUrl points to :18900, restart OpenClaw
401 Unauthorized	Run agentgazer provider add <provider> to store API key
No cost data	Model may not be in pricing table (metrics still captured)
Connection refused	Ensure AgentGazer is running (agentgazer doctor)

Advanced: Manual Configuration

If you prefer to edit ~/.openclaw/openclaw.json manually:

{
  "models": {
    "mode": "merge",
    "providers": {
      "agentgazer": {
        "baseUrl": "http://localhost:18900/agents/openclaw/agentgazer",
        "apiKey": "managed-by-agentgazer",
        "api": "openai-completions",
        "models": [
          { "id": "agentgazer-proxy", "name": "AgentGazer Proxy" }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "agentgazer/agentgazer-proxy"
      }
    }
  }
}

How It Works

1. OpenClaw sends all requests to the agentgazer provider
2. The proxy receives requests at /agents/openclaw/agentgazer
3. AgentGazer looks up Model Override Rules for the agent and routes to the actual provider

Setting Up Model Routing

After applying the config, set up routing in the Dashboard:

1. Go to Agents → openclaw (appears after first request)
2. Click Model Settings
3. For the agentgazer provider entry, configure:
   • Model Override: The actual model to use (e.g., claude-sonnet-4-20250514)
   • Target Provider: The actual provider (e.g., anthropic)

This allows you to change which model/provider OpenClaw uses without editing its config file.

API Key Handling

Set apiKey to any non-empty value (e.g., "managed-by-agentgazer"). The proxy injects the real key stored via agentgazer provider add.

Cost Awareness Skill

When you click Apply Configuration in the Dashboard, AgentGazer automatically installs a cost awareness skill for OpenClaw.

What Gets Installed

The Apply action creates:

~/.openclaw/skills/agentgazer/
├── SKILL.md          # Skill metadata and instructions
└── scripts/
    └── cost.sh       # Script to query AgentGazer stats

Using the Skill

After installation, you can ask OpenClaw about your AI spending:

User: /cost
OpenClaw: Your AgentGazer stats for the last 24 hours:
          - Total cost: $12.45
          - Requests: 847
          - Tokens: 1.2M (in: 800K, out: 400K)

Skill Commands

Command	Description
/cost	Show cost summary for current period
/cost 7d	Show cost for last 7 days
/cost compare	Compare current period vs previous

Manual Installation

If you didn't use Dashboard Apply, manually create the skill:

mkdir -p ~/.openclaw/skills/agentgazer/scripts

# Create SKILL.md
cat > ~/.openclaw/skills/agentgazer/SKILL.md << 'EOF'
---
name: agentgazer
description: Query AgentGazer for cost and usage stats
---

Use the cost.sh script to get spending information.
EOF

# Create cost.sh
cat > ~/.openclaw/skills/agentgazer/scripts/cost.sh << 'EOF'
#!/bin/bash
agentgazer agent openclaw stat -o json
EOF

chmod +x ~/.openclaw/skills/agentgazer/scripts/cost.sh