Skip to main content
ClawHQ gives you visibility into AI spending and tools to keep costs under control — so you can run agents confidently without surprise bills.

​
How costs are tracked

Every model call is logged with:
  • Model used
  • Input and output token counts
  • Estimated cost (based on published pricing)
  • Agent and pipeline that triggered it
View this data in Dashboard → Metrics or via the /api/metrics endpoint.

​
Setting a monthly budget

Set a monthly spending threshold in Settings → General → Budget, or in ~/.openclaw/openclaw.json: 
{
  "modelRouter": {
    "budgetThreshold": 80
  }
}
budgetThreshold is a percentage (0–100). When monthly spend reaches this percentage of your limit, the model router automatically downgrades non-critical tasks to Claude Haiku 4.5 as a fallback.

​
Budget fallback behavior

When the budget threshold is hit:
  • code and research tasks continue using their configured model (they’re considered critical)
  • summary and chat tasks are routed to claude-haiku-4-5 automatically
  • A warning banner appears in the dashboard
You can override this by locking task types: 
{
  "modelRouter": {
    "lockedTaskTypes": ["code", "research", "creative"]
  }
}
Locked task types are never downgraded, even over budget.

​
Cost-saving tips

  • Use Haiku for high-volume, simple tasks (summaries, routing, triage)
  • Use Sonnet for most reasoning tasks — best price/performance ratio
  • Reserve Opus for genuinely complex tasks (architecture, deep analysis)
  • Enable self-learning — the model router learns which models succeed on your tasks and routes accordingly over time
  • Enable Ollama for local routing of simple tasks at zero API cost
See Model Router for full routing configuration.