Controlling industrial equipment via a natural language voice interface is a growing challenge for many manufacturers. Noisy environments can make it difficult to detect voice commands and the complex nature of the control interface for industrial products has so far been difficult to implement securely with natural language methods. However, a new generation of edge AI is using small language models to provide local, native inference specifically for voice control in industrial environments. Industrial engineers often have to work across multiple systems and interfaces to keep production equipment running smoothly. An edge AI agent can help cut through that complexity by removing the need for operators to memorise specialist commands, refer to technical documents or navigate complex interfaces when something goes wrong. As a result, the tool has the potential to reduce training requirements, improve efficiency and make equipment easier and safer to operate. Anritsu, for example, has developed an AI-controlled Vector Network Analyzer (VNA) designed to simplify complex RF, microwave and optical component characterization. The edge AI assistant in the Tensor VNA shows engineers how to set up and execute advanced RF and microwave measurements more efficiently using natural-language guidance. Embedded directly into the instrument, the local AI model acts as a measurement companion, helping engineers solve complex problems and navigate workflows such as amplifier, mixer, S-parameter, calibration and reporting tasks with greater efficiency and confidence. Tensor supports direct measurements up to 1.1 THz with no external test sets required and can streamline amplifier characterization in a single, unified setup, eliminating the need for reconfiguration and minimizing downtime. Key measurements, including gain, compression, AM/AM, AM/PM, THD, and two-tone IMD, can be performed with fewer configuration changes and less measurement uncertainty. The Anritsu Tensor VNA uses a natural language interface to control settings For mixer measurements, the edge AI model simplifies both scalar and vector characterization, including conversion loss, isolation, phase synchronization, group delay, and gain compression across four internal sources, eight receivers and an optional second local oscillator (LO) for efficient configuration of multi-stage mixers and multiple frequency plans. 42 Technology in Cambridge has also developed a natural language AI voice interface demonstrator for industrial applications that is currently being used to control a signal generator using everyday words. “We specifically chose this piece of lab equipment because it was already set up for external control, allowing the team to focus on the LLM, reasoning and hardware-interaction side of the AI agent,” said Mike Sales, Head of AI at 42 Technology. The edge AI allows the signal generator to be controlled using spoken commands via a small language model (SLM) running locally on an Nvidia Jetson single-board computer (SBC). The Jetson SBC handles the complete edge AI workload including speech detection, speech-to-text conversion and agent orchestration. Voice interface is always on at the edge Evaluation kit for neuromorphic voice front end The system combines natural-language understanding with real-time equipment control and machine-state awareness, so users can issue commands, ask questions, and get context-aware support using either spoken or typed instructions. It can quickly retrieve information from a pre-configured library of operator manuals, technical documentation and fault-finding guides to give users immediate contextual help without interrupting their workflow. Unlike consumer voice assistants that run in the cloud, 42T’s system runs entirely on local hardware. This protects sensitive operational data, gives companies complete control over their AI models, and enables constant offline operation without relying on third-party cloud services or subscriptions