Let AI Set It Up for You
You do not have to configure Kilo by hand. Ask the built-in AI assistant, or connect the AI client you already use over MCP, and it does the setup with you.
Most IoT platforms start you with an empty screen and a long form. Kilo does not require that. You can describe what you want in plain language and let AI do the mechanical work — registering devices, writing the automation, wiring up the alarms — while you keep approval over anything consequential.
This page is deliberately the first thing in this section, because the manual walkthrough that follows is not the only way to get started, and for most people it is not the fastest one.
You also do not need hardware to begin. The Emulator connector creates devices that generate their own telemetry from real device presets, so you can build and test a complete deployment — dashboards, rules, escalating alarms — before anything arrives, then point those same devices at real sensors on the day they do.
Two ways to let AI do the work
There are two routes, and they are genuinely different. Pick either, or use both.
Built-in AI Assistant
Your own AI client (MCP)
Where it runs
Inside Kilo, under AI Chat in the sidebar
In the app you already use — Claude Code, Claude Desktop, ChatGPT, Codex, Cursor
Setup needed
None. It is part of the platform
Connect once over MCP and sign in with your normal Kilo account
What it can do
Answer from live and historical telemetry, provision devices and gateways, author, test and deploy rules including their CEL, create alarms with escalation, run device commands, set up emulated devices, manage team roles, recommend hardware
The same operations, exposed as MCP tools, against the same deployment
Permissions
Inherits your exact permissions and organization
Identical — the connection carries your own account's permissions
Confirmation before consequential actions
Enforced by the assistant. Destructive or hard-to-reverse actions surface an explicit Confirm Action / Cancel and do not proceed until you approve
Depends on your client. Kilo marks destructive tools with the standard MCP annotations and states the requirement in the tool description; compatible clients show an approval prompt. Annotations are hints to the host application, not something Kilo can enforce inside third-party software
Best for
Getting set up, day-to-day operation, anyone who wants the platform to walk them through it
Working across your own tooling, scripting, or staying in the editor you already have open
The confirmation row is the one real difference, and it is worth reading twice. Inside Kilo, the gate is ours and it holds. Through an external client, the gate belongs to that client.
What is enforced in both cases: the connection can only ever read and change what your own permissions allow, it stays inside your current organization, a device command must already exist on that device before it can be run, and every dispatch is recorded.
Route 1 — the built-in assistant
Open AI Chat in the sidebar and describe what you want. It already knows your deployment, so you can be direct:
"Register this LoRaWAN sensor — here is the DevEUI and AppKey."
"Alert facilities on-call if any cold store goes above 8 °C for more than 20 minutes."
"Create a test device that behaves like a Dragino distance sensor so I can build the dashboard now."
It will show you what it is about to change and wait for your approval before anything consequential happens. See Building With the Assistant for what it can do and where its limits are.
Route 2 — connect the AI client you already use
If you already work in Claude, ChatGPT, Cursor or Codex, point it at your Kilo organization over MCP and it can operate the same deployment from there. You sign in through the browser with your normal account — there is no API key to mint or paste.
Setup instructions for each client are in MCP Server.
When to do it by hand instead
Doing it manually is still worth your time when you want to understand the mechanics, when you are debugging something specific, or when you simply prefer the forms. Nothing is hidden from you: everything the AI does is an ordinary platform action you could have performed yourself, and it shows up in the same places.
Navigating the Interface tours the platform, and Your First Deployment walks the full LoRaWAN path from gateway to live dashboard.
Last updated