How to Set Up a Custom Model Provider in OpenClaw

OpenClaw supports custom model providers through its Settings panel. This guide walks you through adding ClawPane as a provider so every agent in your gateway gets smart model routing automatically.

Prerequisites

Before starting, you'll need:

• An OpenClaw instance (self-hosted or cloud)
• Admin access to OpenClaw's Settings
• A ClawPane account (free to create at clawpane.co)

Step 1: Create a ClawPane Router

Before adding ClawPane to OpenClaw, you need at least one router:

1. Go to ClawPane Dashboard → Create Router
2. Choose a preset or configure custom weights:
   • Auto — balanced across cost, speed, and quality
   • Fast — lowest latency
   • Economy — cheapest viable model
   • Quality — highest quality regardless of cost
3. Name your router (e.g., production-balanced)
4. Save the router

You can create multiple routers for different workloads later.

Step 2: Get Your API Key

1. Navigate to ClawPane Dashboard → Settings
2. Click Generate API Key
3. Copy the key — it starts with mp_

This key works across all your routers. You only need one.

Step 3: Add ClawPane to OpenClaw

In your OpenClaw instance:

1. Go to Settings → Model Providers
2. Click Add Provider
3. Fill in the fields:

Provider Name:   ClawPane
Provider URL:    https://clawpane.co/route
API Key:         mp_xxxxxxxxxxxxxxxx
Model ID:        auto
SDK:             OpenAI

4. Save the provider

Step 4: Configure Model IDs

The Model ID tells ClawPane which router to use. You have several options:

Model ID	Behavior
auto	Uses your default router (balanced)
fast	Uses the built-in speed-optimized preset
economy	Uses the built-in cost-optimized preset
quality	Uses the built-in quality-optimized preset
your-router-id	Uses a specific custom router

To use different routers for different agents, set the Model ID per agent in OpenClaw's agent configuration.

Step 5: Test the Connection

Create a simple test in OpenClaw:

1. Open any agent that uses ClawPane as its model provider
2. Send a test message
3. Check the response metadata for:
   • x-model-used — which model handled the request
   • x-model-cost — what the request cost
   • x-routing-time — how long routing took

If you see these headers, ClawPane is working correctly.

Troubleshooting

"Connection refused" or timeout

• Verify the Provider URL is exactly https://clawpane.co/route
• Check that your API key is valid and not expired

Agent returns empty responses

• Ensure the Model ID matches a valid router or preset
• Check ClawPane dashboard for error logs

Higher than expected latency

• Routing adds <100ms overhead. If you see more, check your network path
• Try the fast preset to prioritize low-latency models

What Happens Next

Once connected, every request from your OpenClaw agents flows through ClawPane:

1. Agent sends a request via OpenClaw
2. OpenClaw forwards to ClawPane (as a provider)
3. ClawPane scores available models against your router weights
4. Best model is selected; request is forwarded
5. Response returns through the chain with routing metadata

Your agents don't know or care which model handles their requests. The routing is completely transparent.

Advanced: Multiple Routers per Agent Type

For teams with diverse workloads, set up separate routers:

• Support agents → economy router (cost-first)
• Code agents → quality router (quality-first)
• Triage agents → fast router (speed-first)

Each agent in OpenClaw can point to a different ClawPane Model ID, giving you per-workload optimization through a single provider.

Create your first router →