Automotive | Solutions | Liquid AI
Build custom in-car AI assistants that work in real time, on existing hardware.
Liquid's multimodal models run entirely on the CPUs and NPUs already in your production vehicles. Natural voice interaction, cabin vision, and personalized agents are achievable right now, with no cloud dependencies or new hardware.
The challenge for automakers
On their own devices, drivers are accustomed to LLM-grade interfaces. In their cars, they're still getting legacy reactive voice. That's not premium, and it's not the future of luxury. In-car AI is.
Automakers struggle to deploy rich in-vehicle AI assistants because:
Existing large models can't run locally.
LLMs are designed for data centers, not chips behind a dashboard.
Cloud-based processing wasn't made for driving.
Reliability, latency, and privacy concerns are critical for in-car systems.
Existing voice systems are too rigid.
They lack the sophistication to handle natural, multimodal interactions across the cabin.
Our solutions
Edge-first SLMs
SLMs are tuned for low-memory budgets in both CPUs and NPUs. That means sub-second responses for infotainment, navigation, and safety prompts on existing SoCs.
Multimodal-native assistants
Agents are designed from the ground up to process both vision and natural language audio. So they understand intent, emotion, and cabin context, and use it to call hundreds of your vehicle functions.
Hybrid AI agentic architecture
Unique architectures combine the best of edge- and cloud-based AI. You decide which features to run at the edge versus the cloud, and how each operates on your vehicles.
Liquid SLMs can power in-car voice assistants that rival cloud-based AI.
A single model, on existing hardware, running in 20+ languages, at 5x the speed of existing solutions.
Case studies & news
APR 23, 2026Press
Liquid AI and Mercedes-Benz partner to scale embedded in-car intelligence](/content/press/liquid-ai-and-mercedes-benz-partner-to-scale-embedded-in-car-intelligence/index.html)
[08.12Models
**LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge**](/content/blog/lfm2-5-vl-3b/index.html) [AUG 12, 2026Models
**LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge**](/content/blog/lfm2-5-vl-3b/index.html)
[10.01Models
**LFM2-Audio: An End-to-End Audio Foundation Model**](/content/blog/lfm2-audio-an-end-to-end-audio-foundation-model/index.html) [OCT 1, 2025Models
**LFM2-Audio: An End-to-End Audio Foundation Model**](/content/blog/lfm2-audio-an-end-to-end-audio-foundation-model/index.html)
[05.14Case study
**Accelerating Vision-Language Model Deployment for Automotive AI**](/content/use-cases/accelerating-vision-language-model-deployment-for-automotive-ai/index.html) [MAY 14, 2026Case study
**Accelerating Vision-Language Model Deployment for Automotive AI**](/content/use-cases/accelerating-vision-language-model-deployment-for-automotive-ai/index.html)
[05.14Case study
**Scaling Synthetic Video Generation Without Cloud Bottlenecks**](/content/use-cases/scaling-synthetic-video-generation-without-cloud-bottlenecks/index.html) [MAY 14, 2026Case study
**Scaling Synthetic Video Generation Without Cloud Bottlenecks**](/content/use-cases/scaling-synthetic-video-generation-without-cloud-bottlenecks/index.html)