Most "AI" in software is a chat box bolted onto a help page. You ask a question, it returns some text, and nothing in the real world changes. That is a fine demo and a poor operator. The version worth building does the opposite: it is grounded in your actual deployment and it does the work. That is what we mean by AIoT, and it is now built into the core of the Kilo IoT platform.

This post is about what AIoT is, how we approached it (the short version: we were early, we deliberately went quiet, and we spent the time building the infrastructure underneath properly), and the open-source engine we built to make it reliable — Synthetic Brew.

What is AIoT (Artificial Intelligence of Things)?

AIoT — Artificial Intelligence of Things — is the combination of artificial intelligence and the internet of things: AI that runs on the live data flowing off your devices and can act through them, rather than sitting beside them as a separate tool. In practice it means an assistant wired into the platform that knows your deployment end to end — current state, history, and the controls — and works within it.

The distinction that matters is grounding. A general chatbot guesses from training data. An AIoT assistant answers from your real devices, scoped to your permissions, and says so when it cannot retrieve something. That is the difference between something that sounds confident and something that is correct on a live building or fleet.

Beyond a chatbot: what an AI IoT platform actually does

An AI IoT platform earns the name by doing two things, not one.

First, it answers from your real data. Ask "which devices haven't reported in the last 24 hours?" or "what was the average temperature in Warehouse B last week?" and it queries your device history, analyzes the trend, and returns a grounded answer with charts on request — scoped to what you are allowed to see.

Second, and this is the part most tools skip, it does the work. Describe an automation in plain language and it writes the logic, tests it, and deploys it. Ask it to onboard a device and it runs the flow. Ask for an alarm and it builds the escalation chain. You stay in control: before anything consequential it pauses for your explicit confirmation, and once it is done it reads the result back to verify its own work.

How we approached AIoT: the infrastructure before the chat

We were among the first to put an AI chat on top of device data — ask your deployment a question in plain English. It demoed beautifully. Then we deliberately went quiet on AI.

The reason was simple. A chat on top of your data is the easy part. An AI that runs a real building or a real fleet has to be right, and "right" means real context about your specific deployment, scoped permissions, tool access, memory, and a record of everything it does — not a model guessing. So instead of shipping the hype, we spent the time building that infrastructure: an enterprise-grade agent runtime, built in-house. The architecture is here now, and its accuracy grows as the agents are trained on more real-world IoT work.

From answering to acting

Because the assistant is wired into the platform, the same intelligence that answers your question can change something. Tell it the rule you want and it writes the CEL, tests it on a sample payload, and deploys it. And the Rules Engine itself grew hands this release: a rule can now send a command straight to a device the instant its conditions are met. A leak no longer just raises an alarm — it can shut the water off, then alert your team that it happened. Every action is scoped to your access, confirmed before anything irreversible, and recorded in an audit trail. That is the line between a platform that tells you something went wrong and one that does something about it.

Synthetic Brew: the open-source engine behind our AIoT

The hard part of AIoT was never the chat — it was the engine underneath. That engine grew big enough to be a product of its own, so we spun it out and open-sourced it as Synthetic Brew.

Synthetic Brew is a self-hosted AI agent runtime. You describe the agent you need in plain English and it builds, deploys, and orchestrates it — wired to your own tools, knowledge, and memory, kept grounded by retrieval and knowledge graphs so it does not invent answers, running on any LLM you bring, in one Docker command. No lock-in, no per-token markup. It is what powers AIoT inside Kilo, and we were confident enough in the engineering to open-source it so any team can see exactly how it is built or use it to bring reliable AI into their own products faster.

If that is useful to you, the project lives in the open:

GitHub: https://github.com/syntheticinc/syntheticbrew

Website: https://syntheticbrew.ai

What is an IoT platform?

An IoT platform is the software layer that connects your devices, stores and normalizes their data, visualizes it, and lets you automate and alert on it — one place instead of a stack of disconnected tools. AIoT is the next layer on top: the platform that not only holds all of that, but can reason over it and act on it with you.

See it for yourself

AIoT is live in the Kilo IoT platform today. Start free at https://app.kiloiot.io, or if you run multiple sites or don't have an in-house team, book a call at https://kiloiot.io.