AI Models
Reasoning
Whatever the SmoothOperator engineer has to think through — strategy, pace, your questions — is written by a language model from your session’s telemetry: Google Gemini 2.5 Flash by default, or Anthropic’s Claude Haiku 5.5 when we pick it on our gateway (if it fails, Gemini answers). If Google stopped serving 2.5 Flash, the gateway would switch on its own to Gemini 3.6 Flash, with nothing for you to update. Immediate alerts (spotter, flags, incidents) don’t wait for the model: they come from a local bank of phrases.
Your questions are answered by that same model, with the same summary of your race: you press the radio button, speak, and the engineer answers by voice. With a subscription, text and voice go through our gateway, tied to your session or to your activated licence.
Voice (TTS and STT)
By default the engineer speaks with a voice that runs on your own PC, with no network: Piper (a northern English male voice in English; Spanish and Catalan have their own). Immediate alerts play from clips the app downloads once, and anything with numbers in it is synthesised on the spot on your machine. Google Cloud Text-to-Speech is there as an optional network voice and as a fallback.
Your radio questions (push-to-talk) are transcribed by Google Cloud Speech-to-Text, and the answer comes back in the engineer’s voice.
Offline Mode / Free Plan
With no paid plan or API key of your own, or if you prefer not to use cloud AI, the engineer works in “Offline” mode: it spends no AI units and needs no connection while you drive. It is also what it falls back to if you use up your plan’s quota.
It speaks from a bank of pre-generated phrases in the engineer’s voice, downloaded once per language: spotter, flags, incidents, tyres, fuel and gaps, with no network calls. What it does not do is reason or answer your questions: that takes the cloud model.