The Pulse
PATEO Puts Nemotron and Doubao in a Thor-Based In-Car AI Box
PATEO announced an in-car AI Box built on NVIDIA DRIVE AGX Thor, combining NVIDIA Nemotron with ByteDance’s 35-billion-parameter Doubao model. The system runs speech functions locally and uses PATEO’s AI Agent platform to interpret cabin in
AI.info Team · Written with AI · fact-checked ·
Two vehicle programs frame PATEO’s announcement
PATEO’s release, dated Nov. 3, 2026, describes an in-car AI system developed with ByteDance and a leading new-energy vehicle manufacturer at Auto China 2026 in Beijing. The company also says it has received a nomination for a next-generation vehicle project with a leading new-energy vehicle maker. Series production for that program is planned for the second quarter of 2027.
The AI Box is built on NVIDIA DRIVE AGX Thor and combines local AI inference with PATEO’s AI Agent platform. PATEO says the system is designed to understand cabin context, support multilingual interaction and coordinate vehicle functions. The announcement presents the box as a dedicated computing addition, rather than a replacement for a vehicle’s original main control chip.
Nemotron and Doubao handle speech and language
PATEO integrates NVIDIA Nemotron 3.5 ASR for speech recognition in its embedded-model suite. It has also integrated ByteDance’s Doubao, a mixture-of-experts language model with 35 billion parameters. The suite includes text-to-speech and a two-billion-parameter language model for semantic classification and rejecting requests.
PATEO says the integrated suite can run inference locally without a cloud connection, supporting multilingual voice interaction and real-time streaming speech recognition. The company says that keeps core AI functions available when connectivity is limited, while connected services and cloud updates can extend the system’s capabilities. Its AI Agent platform combines voice, visual and sensor inputs to interpret user requests and cabin conditions.
Within the vehicle’s safety permissions, the agent can invoke cabin controls and online services, coordinate simultaneous tasks and handle requests that run for a longer time. With user consent, it can retain preferences and habits to personalize later interactions. PATEO’s release says the system can combine those retained preferences with cloud updates that refine its functions.
Cabin prompts range from cooling to fatigue
PATEO says its Sentinel AI can flag parking violations, floods and fires. In hot weather, occupants can adjust the driver’s and passenger’s air conditioning separately by voice command or gesture. On long highway trips, the agent can detect driver fatigue and ask whether the driver wants to open a window or activate the massage seat.
In areas with poor air quality, the system can suggest closing the windows and switching to recirculation mode, using location and learned driving habits. These examples combine cabin controls with voice, gesture or other inputs, while the release says vehicle safety permissions govern the agent’s access to controls. The system can also coordinate online services alongside in-car functions.
Thor’s memory bandwidth matters to local inference
PATEO says supporting these functions requires memory bandwidth as well as computing power measured in TOPS. As conversation context and agent actions accumulate during a trip, the model’s key-value cache grows, increasing demands for memory capacity and bandwidth. The company says Thor’s accelerated computing and high memory bandwidth contribute to responsive local inference.
The AI Box’s modular software architecture connects with embedded systems and NVIDIA’s edge and cloud AI ecosystem to support model deployment, tuning and customization across vehicle programs. PATEO says adding dedicated AI computing can provide an upgrade path for existing vehicles whose processors cannot support larger models, simultaneous multimodal interactions or proactive agents through software updates. The design leaves the original main control chip in place.
PATEO links the box to recurring services and robotics
PATEO describes its embedded tokenomics strategy as a way to generate recurring revenue from deploying and using AI capabilities, extending business opportunities beyond the initial hardware sale. In June 2026, the company formed partnerships with Mininglamp Technology, Xunce Technology and Saimo Technology to develop a token-based physical-world AI model and TokenOS. The collaboration brings together embedded hardware, simulation testing and data infrastructure, according to the release.
PATEO also says it received a project nomination from a major domestic provider of robotic computing solutions. The company says it is extending its work in intelligent interaction and hardware engineering into robotics, applying its platform beyond vehicles. Its AI Box is described as a reusable platform that PATEO can adapt for different automakers and vehicle programs.