ETS shows you what you programmed. Not what it does.
Drop the project file in. You get back what the installation is actually doing — the parameters inside the devices, the automation already running on the bus, and the defects nobody wrote down.
Free, permanently. No card. You do not talk to anyone unless you want to.
| Room | Address | Hold |
|---|---|---|
| Garage | 1/1/8 | 120 s |
| Hall | 1/1/2 | 90 s |
| Living | 1/1/5 | 60 s |
| Kitchen | 1/1/12 | 45 s |
| Study | 1/1/31 | 10 s≥ 30 s |
| Bath 1 | 1/2/4 | 10 s≥ 30 s |
| Bath 2 | 1/2/9 | 10 s≥ 30 s |
| Laundry | 1/2/14 | 10 s≥ 30 s |
One real project, read in full
Every figure below is counted, not estimated. We publish no saving, no percentage and no payback figure, because we cannot measure one.
| Qty | Item | Note |
|---|---|---|
| 668 | Group addresses | The complete address map |
| 234 | Addressed, wired to nothing | Declared, never terminated |
| 113 | Addressing issues | Conflicts and orphans |
| 1403 | Device parameters resolved | 100% — hold times, staircase timers, thresholds, scene tables |
| 21 | Automations already running | On the devices themselves. In no documentation |
| 12 | …of those, sensor-driven | Acting without anyone knowing |
| 25 | Recommendations | Each traced to the parameter that produced it |
| 4959 | Owner edits since handover | From the project's own ETS log, since 2026-02-15 |
Disclosure. The reference installation is a house in Lindfield, NSW 2070 — the author's own house. It is the only installation we have analysed in full, and we would rather say so than imply a client base we do not have.
A tool for Tuesday, not a report you read once
The findings are the start. What follows is the part that saves the day: fix what you agree with, put the file back in, and see precisely what moved.
- 01UploadA .knxproj. About a minute.
- 02ReadAddresses, devices, topology, and every parameter the products declare.
- 03AskA model trained on the devices in your project — not general advice about KNX.
- 04WorkGroup addresses, devices and device-local behaviour in one workspace, beside ETS rather than instead of it.
- 05FixTake the recommendations into ETS. Each names the parameter that produced it.
- 06DiffPut the new file in. Two revisions compared: what changed, what did not, what is still open.
- 07Hand overA handover document for your client, generated from the project, under your name.
- 08KeepThe project stays in your workspace. When they ring in eighteen months, it is a search.
Registration, no card. Exports are included during the trial. The check itself stays free — that is a decision, not an introductory offer.
Where the bus stops
Not because KNX is limited — because a device acts on what it alone can sense. That is a wiring fact, not an opinion about one system being better than another.
A device decides this on its own
- —Presence in the room it is in
- —Lux at its own sensor
- —A staircase timer it holds itself
- —A scene it was told to store
These need two systems at once
- ×Occupancy across rooms
- ×A tariff
- ×A forecast
- ×A calendar
- ×A person's phone
Everything in the right-hand column is what the next section is about.
A house that decides, and writes down why
To read a project properly you need a model of a house: what a room is, what a device can do, what depends on what. We spent two years building one, and it is the reason the check finds things — and the reason software can be trusted to make decisions in a house continuously, because a guardrail has to be written against something.
The Casient appliance is one box: a KNX bus interface, a bus monitor, that model, a policy engine, the guardrails, and an AI layer that proposes. Home Assistant runs inside it, hardened — integrating two thousand products is a solved problem and we did not re-solve it. The operating system is kept separate from the automation platform, so the monitoring is still running while the automation is being changed.
Policies propose. An arbiter resolves. Every action is written down — with the policy that won, the ones it beat, the evidence, and the tier it resolved on.
Anything that changes the boundary of the house needs two independent sources. Two readings from the same detector count as one. A refused action is still a line in the log.
Primary reflexes stay on the bus. Staircase timers, motion holds, scene tables and central-off keep running on the actuators. Lose the box, the network and the internet, and the house still behaves.
A Basalte Core, a Logic Machine, a Comfortclick Jigsaw, a 1Home server all do a real job and do it well: they bridge a bus to an interface and run rules somebody wrote. If that is what your project needs, buy one. They are mature, they are supported, and we are not yet.
What none of them does — because it is not what they are for — is decide. A rule does not learn that this house goes to bed at eleven except on Fridays, and a rule that has never been wrong has also never been audited.
This is in beta, in one house, which is ours. We are looking for the second one. If you think something here is wrong, or already exists, tell us — we would rather be corrected than flattered.
Decisions we wrote down before we needed them
These are architecture decisions, not slogans. Each one is recorded and dated, and each one rules out something we might otherwise have built.
- Nothing leaves the house unless the owner opens the door.
- We cannot let ourselves in. Support access is granted by the owner, for a set time, and every session is logged.
- We keep a written list of what we are allowed to store. Anything not on that list is refused.
- A break-in at our end cannot reach a house. The box checks signatures itself and rejects anything it cannot verify.
- Australian homes' data stays in Australia.
- Backups are encrypted and kept off the box. A replacement box is running again within the hour.
- The house keeps working without us. Safety and comfort run on the bus, not on our software.
- Every decision is written down: what was chosen, what it beat, and why.
Every job you finish stops earning
You install for weeks and invoice once. The client's expectations keep moving, because they came from a phone. Your competitor quotes the same hardware you do.
Time
Reading a project you did not commission takes hours. The check takes about a minute, and the handover pack writes itself.
Something to say
A finding quoted from the manufacturer's own manual is not an opinion. It changes what you can tell a client, and what you can charge to fix.
More from the same equipment
The sensing and the actuation are already in the wall. What is missing is anything able to use two systems at once.
Revenue after handover
Monitoring, backup, support and improvement — in your name. White-labelled, with a fleet view across every house you run.
The client stays yours. The handover pack goes out under your name; ours appears only if you decide it should. We do not market to your clients, and when one of them contacts us directly, we tell them to call you.
One person, one house, and a project file that would not stop being wrong
I am Ted, and Casient is mine. The house in the schedule above is where I live. I commissioned it, I have had ETS open at eleven at night, and I built the check because I wanted to know what my own project was actually doing and nothing would tell me.
There is no sales team. If you write, I answer.
This is for a specific person
You commission four or more KNX projects a year.
You get asked to connect KNX to a control system — an app, Home Assistant, a visualisation — and it is painful every time.
You have a project attached right now, not a general interest.
Twenty-five projects a year with no integration work is not a problem we solve. If that is you, we are probably not what you need yet — and the check stays free regardless, so run it anyway if you are curious.
Drop your .knxproj in and see what it says about your project.
Not a demo. Not a call. The file goes in, the findings come out, and you decide whether any of it is worth a conversation.
If it is the rest of it you want to talk about,write to Ted. There is no form to fill in first.