Let OpenRouter Pick the Model for You | Field Notes

Why pay Claude Opus prices to process a heartbeat? You shouldn't.

Your agent does dozens of things daily. Some need serious reasoning. Others are trivial — checking a file, responding HEARTBEAT_OK, parsing JSON. Running everything through a premium model is burning money on capability you don't need.

Enter OpenRouter Auto

openrouter/auto analyzes each prompt and routes it to the most cost-effective model that can handle it well. No manual config needed.

Heartbeats — checking HEARTBEAT.md → cheap model
File reads — extracting a config value → cheap model
Simple questions — "What time in Tokyo?" → cheap model
Complex tasks — writing code, debugging → premium model

Your bill drops without you micromanaging model selection.

Setup

In your openclaw.json:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/openrouter/auto"
      },
      "models": {
        "openrouter/openrouter/auto": {}
      }
    }
  }
}

That's it. Add a fallback for extra reliability:

{
  "model": {
    "primary": "openrouter/openrouter/auto",
    "fallbacks": [
      "openrouter/anthropic/claude-haiku-3.5"
    ]
  }
}

When Auto Makes Sense

Ideal for:

24/7 agents with lots of background tasks
Cost-conscious setups wanting automatic optimization
Variable workloads — sometimes simple, sometimes complex

Less ideal when:

You need guaranteed behavior from a specific model
Model consistency matters for sensitive work
You've already manually optimized model routing

The Savings

A typical agent processes 50+ heartbeats daily plus routine checks. With Auto, those route to models costing 1/50th as much. Complex work still gets top-tier models.

Users report 40-60% reductions without noticeable quality drops on actual work.

Bottom Line

Set Auto as your primary, run for a week, check your dashboard. Let the router do the work. Save your money for the tasks that actually need it.