xZane runtime puts safety first between AI and hardware
New open-source runtime xZane sits between your LLM and robot, validating every command before it executes.
Julio Chinchilla has open-sourced xZane, a runtime designed to bridge the gap between large language models (LLMs) and physical actuation while enforcing safety. Unlike traditional middleware that only pre-validates commands, xZane introduces an independent safety layer that runs alongside execution. This ensures device-specific safety functions—such as certified emergency stops—remain uncompromised, even when the LLM proposes actions.
The runtime has been tested across three diverse devices: a ROS 2 mobile robot (Yahboom Rosmaster M1 with Jetson Orin Nano), an LG webOS TV, and a robot vacuum. Chinchilla validated it using multiple LLMs, including cloud-based models like Claude, GPT, and Gemini, as well as local options like Mistral and Qwen. By using a schema to define device capabilities rather than hardcoding logic, the same runtime can adapt to different hardware without code changes. The approach positions xZane as a flexible solution for robotics, industrial automation, and smart home devices where safety and reliability are paramount.
- xZane is an open-source runtime that validates AI-generated commands against device capabilities before execution, ensuring safety.
- Tested on three devices (ROS 2 robot, LG TV, robot vacuum) using multiple LLMs (Claude, GPT, Mistral, etc.).
- Runs safety checks independently of the AI model, integrating with existing safety functions like e-stops.
Why It Matters
Solves a critical gap in AI-driven robotics and automation by ensuring safe, schema-based validation without sacrificing flexibility.