As large AI models accelerate their integration into vehicles, the competitive dynamics of intelligent cars are changing. In the past, smart cockpits largely focused on voice assistants, in-car applications, and multimedia services. Today, with on-device omni-models, AI agents, and AI operating systems gradually becoming reality, cars are evolving from smart terminals that execute commands into AI agents capable of understanding users and proactively providing services.
Recently, TechNode had the opportunity to speak with the Banma Intelligence team about AI-powered cockpits, on-device omni-models, AI agents, and the globalization of Chinaโs intelligent automotive software industry.
The company describes itself not only as a provider of AI models for smart cockpits, but also as a platform technology company integrating full-stack AI technologies, service ecosystems, and engineering capabilities. At the core of this strategy is its vision of AI in All.
Founded in 2015, Banma Intelligence has evolved from perception-based AI to generative AI and now agentic AI. From its early AI voice assistant to the launch of its AI in All strategy and the Yan AI smart cockpit technology brand in 2024, the company has been gradually building a full-stack AI technology system spanning foundation models, AI operating systems, and AI agents.
From smart cockpits to AI cockpits
Banma Intelligenceโs current AI cockpit strategy is centered on Yan AI, covering foundation models, on-device models, AI agents, and AI-native operating systems.
Yan AI is a suite of models optimized for smart cockpit scenarios. Leading the family is AutoOmni, an on-device, omni-modal foundation model capable of integrating visual, voice, and textual information for continuous perception and understanding of users.
This approach is reflected in Banma Intelligenceโs concept of No Touch, No App. In the past, users typically had to activate a voice assistant, issue a command, or navigate through the in-car system to find a specific function. Banma Intelligence envisions vehicles that can understand users proactively, rather than waiting for them to take each step.
For instance, a vehicle could assess a driverโs level of fatigue and proactively adjust the seat and music, or use the userโs schedule to plan routes and charging stops in advance. In this model, the vehicle shifts from simply executing commands to proactively providing services.
AI agents take cars from answering questions to getting things done
Beyond the models themselves, AI agents are becoming a key focus for Banma Intelligenceโs next stage of development.
Last year, Banma Intelligence launched SystemAgent, which marked a shift from conversational AI toward task execution through a SystemAgent + AI Agents architecture. The company then introduced SuperAgent to bring agent capabilities into a broader range of scenarios, including entertainment, mobility, lifestyle, and vehicle services.
This year, the company further launched AutoClaw, designed to break down complex user requests, plan and iterate on tasks, and call on different agents and tools to complete them. This marks a shift in the value of automotive AIโfrom simply answering questions to getting things done on behalf of users.
In a parking lot, a parking assistant agent could use exterior cameras to gather information about the surroundings, while an on-device foundation model handles recognition and decision-making before calling relevant services to complete parking payments. When users leave the vehicle, the system could also identify items such as smartphones, laptops, and water bottles and proactively remind them if anything has been left behind.
As these capabilities mature, cars could evolve beyond terminals that provide information and entertainment into AI agents capable of understanding their surroundings, calling tools, and executing tasks.
From SDV to AIDV
In Banma Intelligenceโs view, these changes ultimately point to another shift in the competitive logic of the automotive industryโfrom Software-Defined Vehicles (SDV) to AI-Defined Vehicles (AIDV).
In the SDV era, automotive competition has expanded beyond traditional hardware capabilities such as engines and chassis to include software capabilities such as smart cockpits, intelligent driving, and over-the-air updates. AIDV takes this evolution a step further by embedding AI deeper into the underlying architecture of vehicles.
Banma Intelligence believes that a truly AI-native vehicle is not simply a conventional car with an additional AI feature. Instead, AI needs to serve as part of the underlying operating system and core decision-making layer, with the vehicle designed around AI across its architecture, data, interaction, and driving systems.
This makes AIOS a potentially critical piece of infrastructure for the next stage of development. Banma Intelligence is building a technology stack spanning chip adaptation, system infrastructure, on-device omni-models, and an AI agent ecosystem. Its goal is to make AI more than an application running on top of the automotive operating system, and instead turn it into part of the vehicleโs digital foundation.
On-device AI and global expansion
Between cloud-based and on-device AI, Banma Intelligence sees long-term value in on-device AI for vehicles. Cars need to continuously process large amounts of real-time information from cameras, microphones, and vehicle sensors, while also handling privacy and safety concerns. This makes low latency, offline availability, and data security important requirements for automotive AI.
However, Banma Intelligence does not see cloud and on-device AI as mutually exclusive. Instead, it is pursuing a cloud-device collaborative approach: complex tasks can leverage cloud computing, while tasks requiring real-time responses and stronger privacy protection can be handled primarily on the vehicle.
At the same time, the development of automotive AI is creating new opportunities for Chinese intelligent automotive software companies to expand globally.
Banma Intelligence has worked with 69 automakers to date, with its solutions deployed in more than 10 million intelligent vehicles, according to the company. It has also adapted its technology to around 10 chip companies and over 30 chip platforms.
For global expansion, the company is pursuing a dual-track strategy of In China for Global and From China to Global. This includes working with international brands such as Volkswagen and BMW, while also supporting Chinese brands including IM, Roewe, MG, and Jetta in overseas markets, bringing capabilities developed in China to global markets.
For Chinese automotive software companies, however, true globalization is about more than replicating products in overseas markets. It also requires localized adaptation, data compliance, cybersecurity, and the development of local service ecosystems.
