AI Advisor

Overview

A question goes to a large language model, which decides what station data it needs and asks for it through tools. Live values, histories, alarms, the audit log, the navigation tree, and tag queries are all in reach, so answers are about the actual site, not about buildings in general. The tools only read: it cannot command a point or change configuration. Provider and model are chosen from a dropdown: OpenAI, Anthropic, Google, Azure OpenAI to keep traffic inside your own tenant, or a local Ollama server for sites that will not send anything off site. The API key is yours, so tokens are billed by the provider and the licence covers software alone. Token use is tracked in the station. Pipelines are small wire-sheet programs that run on demand or on a schedule. Three are ready to use: forgotten manual overrides, alarm patterns over the last day, and Niagara network health. Installs on a Supervisor or Workbench host, Windows or Linux, and is licensed by subscription.

List of FAQs

Do I need an account with an AI provider? Yes, you create an account with OpenAI, Anthropic, or Google, paste the API key into the station, and the tokens your questions consume are billed by that provider directly. Everyday use is cheap, a simple question costing a fraction of a cent, and the station tracks cumulative token use so spend stays visible. Does our building data leave site? With a cloud provider, yes, and it is worth being clear about it: the question and whatever station data the model asks for travel to that provider over HTTPS. For sites where that is not acceptable, point it at a local Ollama server instead and nothing leaves your network. Azure OpenAI is a middle path that keeps traffic inside your own Azure tenant. Can it change anything in the station? No. Every tool it has is a read: point values, histories, alarms, the audit log, the navigation tree, and tag queries. It cannot command a point, write a setpoint, or alter configuration. The single exception is a file it writes into its own folder to store the station profile it builds when first set up. What sort of questions does it handle well? Ones that would otherwise mean opening several views and comparing them by hand. Why a unit is not holding setpoint, which alarms have repeated this week, what a value was doing overnight, which devices are offline. It reads the same data an engineer would, just faster. What are pipelines? Automated checks rather than conversation. A pipeline pulls data on a schedule, sends it for analysis, and posts the outcome to a dashboard as a red, amber, or green card, so problems surface without anyone asking. Three come ready made, covering manual overrides left on, alarm patterns over the last day, and Niagara network health, and you can build your own on a wire sheet. Which providers and models can I use? OpenAI, Anthropic, Google, Azure OpenAI, and Ollama for local models.