PATEO announced its AI Box on November 3, 2026, saying its in-car AI suite can run inference locally without a cloud connection. The system is built on NVIDIA DRIVE AGX Thor and adds dedicated AI computing alongside a vehicle’s original main control chip, according to PATEO’s announcement.
The models in PATEO’s AI Box
PATEO names NVIDIA Nemotron 3.5 ASR for speech recognition and ByteDance Doubao, a 35-billion-parameter mixture-of-experts language model. The suite also includes text-to-speech and a separate, unnamed 2-billion-parameter language model for semantic classification and request rejection.
That smaller model has a distinct stated role: it classifies meaning and screens requests, alongside the speech-recognition and language-generation components.
What PATEO says works offline
PATEO says the suite can perform local inference without a cloud connection, with core AI functions available under limited connectivity. The company describes connected services and cloud updates as extensions of the system’s capabilities.
The announcement does not assign offline operation feature by feature. Its claim is about local processing by the AI suite; PATEO separately describes connected services as part of the system.
Described cabin functions and production timing
PATEO says its AI Agent combines voice, visual and sensor inputs, and can invoke cabin controls and online services within vehicle safety permissions. Examples in the company’s description include Sentinel AI alerts for risks such as parking violations, flooding and fire; separate climate adjustments for the driver and passenger by voice or gesture; and fatigue-related prompts during extended highway driving. PATEO also says preference memory requires user authorization.
PATEO reports that an unnamed new-energy vehicle automaker nominated the system for a next-generation vehicle program. The company expects mass production for that program in Q2 2027.
