OpenRouter + OpenClaw: Access 300+ Models in 2 Minutes

OpenRouter + OpenClaw: Access 300+ Models in 2 Minutes

Most people set up OpenClaw with one provider. Anthropic. Maybe OpenAI. Then they're stuck.

Want to try Llama 3 for a coding task? New API key. New config. New billing dashboard. Want to benchmark Claude vs GPT-4 vs Gemini on the same prompt? Three accounts. Three keys. Three invoices.

OpenRouter fixes this. One API key. 300+ models. 60+ providers. Same OpenClaw setup.

Here's how to wire it up.

Why OpenRouter for OpenClaw

Three reasons this matters:

1. One key, every model. Claude Sonnet, GPT-4.1, Gemini Flash, Llama 3.1, Mistral, DeepSeek — all through a single sk-or- key. No juggling provider dashboards.
2. Free tiers exist. Some models on OpenRouter have free usage tiers. If you're building a PoC or just learning OpenClaw, you can experiment without burning credits.
3. Fallback routing. If one provider goes down, OpenRouter can route to another. Your agents stay up.

If you've been locked into a single model because switching providers felt like a chore — this is the fix.

Prerequisites

You need three things:

1. OpenRouter account — Sign up at openrouter.ai
2. API key — Generate one at openrouter.ai/settings/keys
3. OpenClaw running — Either self-hosted or via Clawctl

Your API key will look like this: sk-or-v1-abc123...

Keep it safe. Treat it like a password.

Step 1: Get Your API Key

Go to openrouter.ai/settings/keys.

Click Create Key. Give it a name like "openclaw-production" so you remember what it's for. Copy the key — you won't see it again.

Add credits. Even $5 is enough to get started. Some models (like certain Llama and Mistral variants) have free tiers, but most production models charge per token.

Step 2: Configure OpenClaw

Add OpenRouter as a provider in your OpenClaw config:

Option A: Environment variable (recommended)

export OPENROUTER_API_KEY="sk-or-v1-your-key-here"

Option B: Configuration file

In your openclaw.json:

{
  "credentials": {
    "openrouter": {
      "apiKey": "sk-or-v1-your-key-here"
    }
  }
}

Don't commit your key to git. Use environment variables or a secrets manager in production.

Step 3: Pick Your Model

This is where it gets fun. OpenRouter gives you access to models from every major provider — through one API.

Configure your agent's model:

{
  "agents": {
    "list": [
      {
        "id": "main",
        "model": "openrouter/anthropic/claude-sonnet-4-5",
        "workspace": "/path/to/workspace"
      }
    ]
  }
}

Top model picks for OpenClaw agents:

The model ID format is provider/model-name. Check the full list at openrouter.ai/models.

Recommendation: Start with anthropic/claude-sonnet-4-5. It's the best all-rounder for agentic work. Swap to cheaper models for low-stakes tasks once you know your workload.

Step 4: Test the Connection

Fire up OpenClaw and verify everything works:

openclaw start

Test with a simple prompt:

openclaw chat "What model are you? Tell me your exact model ID."

You should see a response confirming the model you configured. If you picked Claude Sonnet through OpenRouter, it'll behave exactly like native Anthropic — because it IS Claude, just routed through OpenRouter's API.

Step 5: Set Up Cost Controls

OpenRouter charges per token, just like direct provider APIs. Set limits so you don't wake up to a surprise bill:

{
  "agents": {
    "list": [
      {
        "id": "main",
        "model": "openrouter/anthropic/claude-sonnet-4-5",
        "tokenLimits": {
          "maxInputTokens": 100000,
          "maxOutputTokens": 4096,
          "maxTokensPerDay": 500000
        }
      }
    ]
  }
}

You can also set spending limits directly on OpenRouter's dashboard under Settings > Limits. Belt and suspenders.

Step 6: Multi-Model Setup (Advanced)

Here's the real power move. Run different models for different agents:

{
  "agents": {
    "list": [
      {
        "id": "analyst",
        "model": "openrouter/anthropic/claude-sonnet-4-5",
        "systemPrompt": "You analyze data and produce reports. Be thorough and precise."
      },
      {
        "id": "web-scraper",
        "model": "openrouter/google/gemini-2.0-flash-exp",
        "systemPrompt": "You process web data quickly. Prioritize speed over depth."
      }
    ]
  }
}

Why this works:

• Analyst agent uses Claude Sonnet — better at reasoning, worth the extra cost
• Web scraper uses Gemini Flash — fast, cheap, good enough for data processing
• One API key manages both. One bill. One dashboard.

This is the setup pattern that lets you build composable multi-agent systems without managing three different provider accounts.

Step 7: Production Security

Lock it down before deploying:

{
  "gateway": {
    "host": "127.0.0.1",
    "port": 3000,
    "authToken": "your-secure-gateway-token"
  },
  "security": {
    "egressControl": {
      "enabled": true,
      "allowedDomains": [
        "openrouter.ai"
      ]
    }
  }
}

Egress control matters. Your agent only needs to talk to openrouter.ai. Block everything else. This prevents data exfiltration if an agent gets a malicious prompt.

Common Issues

"Invalid API key" Error

Check that:

1. Your key starts with sk-or- (not sk-ant- or sk-)
2. The key hasn't been revoked on OpenRouter's dashboard
3. You have credits on your OpenRouter account

"Model not found" Error

OpenRouter model IDs use the format provider/model-name:

• ✅ openrouter/anthropic/claude-sonnet-4-5
• ❌ claude-sonnet-4-5
• ❌ anthropic/claude-sonnet

Check openrouter.ai/models for exact model IDs.

Slow Responses

Some models on OpenRouter are hosted by third parties with variable latency. If speed matters:

• Use google/gemini-2.0-flash-exp for fastest responses
• Stick to top-tier providers (Anthropic, OpenAI, Google) for consistent latency
• Check the "Latency" column on OpenRouter's model page

Free Tier Rate Limits

Free models have lower rate limits. If you're hitting limits:

• Add $5-10 in credits and use paid models
• Or reduce request frequency with token limits

Cost Comparison

Same model, different routes:

OpenRouter doesn't mark up most major models. You pay the same price but get the flexibility to switch models without changing your setup.

Budget-friendly alternatives on OpenRouter:

With Clawctl

If you're on Clawctl, the setup is even simpler. No config files.

1. Open your Dashboard
2. Click the Setup Wizard
3. Select OpenRouter as your provider
4. Paste your sk-or- API key
5. Clawctl validates, encrypts, and deploys

That's it. Your key is stored encrypted (AES-256-GCM), injected into your agent at runtime, and never exposed in logs or API responses.

Clawctl also handles:

• Egress control — Only openrouter.ai is allowed by default
• Human approvals — 70+ high-risk actions require explicit approval
• Audit trail — Full searchable log of every agent action
• Cost monitoring — Token usage tracking with alerts

You get the model flexibility of OpenRouter with the production security of Clawctl. Best of both.

Set up OpenRouter on Clawctl →

What's Next

Once you're running OpenRouter + OpenClaw:

• Benchmark models — Try the same prompt across Claude, GPT-4, and Gemini. Pick the best fit for each task.
• Build multi-agent setups — Use Claude for reasoning, Gemini for speed, Llama for privacy-sensitive tasks.
• Explore free tiers — Test new models without spending a dime.

300+ models. One key. Go build something.

Full OpenClaw documentation →