Pricing

Questions people ask

What counts as a decision?

One routing choice. Your agent makes a call, the router picks a model, and that single choice is one decision — whatever the model then costs you.

Why does this cost twice what RealitySignal alone costs?

RealitySignal on its own makes your existing decisions better. RealityRouter makes the decision and picks the model — which is the part that cuts your inference bill. On a ten-task SWE benchmark, routing solved more tasks than GPT-6 Astra for about half the money. If you only want the calibration, RealitySignal is $99.

Is the router really free?

Yes. MIT licence, source on GitHub, self-hosted, unlimited. If you never pay us anything, it keeps routing — it just routes on uncalibrated estimates.

Do you see our prompts?

No. Prompts go from your infrastructure straight to your provider. Our calibration service receives anonymised task features only. Choose self-hosted Enterprise and even that stays inside your network.

Do you take a percentage of our inference spend?

No, and we never will. You pay your providers directly with your own keys.

Can we use local models?

Yes. Any OpenAI-compatible endpoint, including Ollama and anything you host yourself. Local models are the majority of routed traffic on our own deployments.

What happens if we stop paying?

The router is MIT and keeps working. You lose calibration, not your infrastructure.