State Labs provides an edge‑native AI platform that runs and fine‑tunes small vision‑language models on consumer, industrial and robotic devices with up to 90 % lower compute requirements, keeping raw data on‑device for privacy. The solution includes a governable model framework that tracks contribution provenance and enforces usage rights, and offers hardware‑agnostic SDK and APIs for OEMs and robotics manufacturers to embed low‑latency intelligence directly into their products.
Funding
Funding not disclosed
Founders
Product
Problem
Deploying large language models on edge devices and robotic platforms incurs high compute costs and latency, making real‑time inference impractical. At the same time, transmitting raw user data to cloud services raises privacy and safety concerns, especially in high‑risk industrial environments. Existing solutions also lack mechanisms for shared ownership and governance of AI models across multiple contributors.
Solution
State Labs provides an edge‑native AI substrate that delivers inference and on‑device training for sVL/sVLA models with up to 90 % lower compute requirements. The platform keeps raw data local, ensuring privacy by design while reducing bandwidth dependencies. A governable model framework enables contributors to retain ownership and define usage rights across datasets and training efforts. The stack supports consumer‑grade devices, industrial service endpoints, and robotic control systems, delivering low‑latency, robust performance in real‑world deployments. Integration is exposed through a lightweight SDK and hardware‑agnostic APIs, allowing OEMs to embed collaborative intelligence directly into their products. Continuous licensing and custom integration services give partners rapid time‑to‑market while maintaining control over their AI assets.
Target Audience
Primary customers are hardware OEMs, robotics manufacturers, and industrial IoT service providers seeking on‑device LLM capabilities with built‑in privacy and governance. The solution also serves consumer device makers that require instant, offline AI responses.
Features
- Edge‑optimized inference engine achieving up to 90 % reduction in FLOPs for sVL/sVLA models
- On‑device fine‑tuning capability that eliminates the need for cloud‑based training loops
- Privacy‑first data handling: all raw inputs remain on the device with end‑to‑end encryption
- Governable AI layer that tracks contribution provenance and enforces shared ownership policies
- Unified VLM/VLA stack tailored for low‑latency robotic perception and control
- Cross‑platform SDK with support for RISC‑V, ARM, and custom ASIC architectures
- Plug‑and‑play integration APIs for consumer electronics, industrial IoT gateways, and autonomous robots
- Licensing model with optional custom integration services for hardware manufacturers and robotics firms