Home › Help guides
How it works — access, safety and what happens to your system
Updated 25 September 2026
Letting someone into your Home Assistant is a real decision. This page explains exactly what access we get, what we do with it, and how it ends, in plain English. If anything here doesn't match what you see on your job page, tell us: that would be a fault.
How we connect
There are two ways, and your job page picks the one that suits your setup.
The ElectriFix Connect integration (no token, no open ports). You add a small integration to your Home Assistant through HACS and paste in a job code from your job page. Your Home Assistant then makes one outgoing connection to us, the same kind your browser makes to any website. Nothing is opened up at your end, and nobody can dial in.
- The integration creates its own access token inside your Home Assistant. You never make one and never paste one anywhere. That token never leaves your Home Assistant; we never receive it and never store it.
- The integration makes only a short, fixed list of changes, and refuses every other change inside your house, before it reaches Home Assistant. It never reads camera images.
- The token shows in your profile as ElectriFix Connect, so you can see it and revoke it yourself at any time. The integration revokes it automatically when the job finishes, or the moment you disable or remove the integration.
A token you create (if your Home Assistant is already reachable from the internet). If you already reach your Home Assistant from outside the house, through Home Assistant Cloud or your own address, you can instead create a long-lived access token in your profile and paste it into your job page. We keep it encrypted for the life of the job and destroy our copy when the job closes. The token itself lives in your Home Assistant, so delete it from your profile when you're done; it stops working the moment you do.
We never ask for your Home Assistant password.
What happens in a job
- We check we can reach you, before you pay. If we can't connect, you aren't charged.
- You tell us what's wrong. In your own words, if you want to. The report starts with what you asked about.
- We read your whole setup. Entities, devices, automations and scripts, integrations, the log, repair notices, the history database and your backups. This read is read-only: it cannot change anything.
- We work out what's wrong. Findings come from fixed rules in code, not guesswork, so the same system gives the same findings every time. You get them in plain English.
- We propose a plan. The plan is drafted by an AI model, Claude, made by Anthropic and reached through OpenRouter; both are in the United States, so what the model is sent leaves Australia (our privacy policy lists exactly what that is). It works from those findings, and every proposed change is then checked by fixed rules in code before you see it. A change the rules can't prove is justified from what we read is refused, not shown to you. Your access token is never sent to the model.
- You tick what you want. Every change is a line on a tick-list, in plain English. Nothing is applied that you didn't tick, except one backup: it's offered first and starts ticked, and if you approve a change that can only be undone by restoring (an automation rewrite, a registry removal, trimming history), the backup is taken first even if you unticked it. Anything that can't be undone is labelled that way and is never pre-ticked. You can tick nothing at all; that's a real answer.
- Each change goes through the same safety cycle. Before a change, we record how things are. We make the one change, wait for Home Assistant to settle, and compare. If something changed that the change shouldn't have touched, it's undone automatically. If an undo ever fails, the job stops there and nothing more is changed. A person will look at this within three business days.
- You get a written report of everything we found, everything we changed, and everything we didn't and why.
- You can ask for more. Your tier includes follow-up rounds: "fix this, delete those, add that" counts as one. Each gets the same read, plan, tick and safety cycle.
- Access ends. When you say you're done, or when your seven days are up, the job closes and access is removed as described above.
What the rules check
Every kind of change has its own rule, and each of our fault guides states the rule for that fault. A few that apply to everything:
- A change must be justified by something the read actually found. We don't "tidy up" things that aren't broken.
- Anything that references a device must reference one that exists and is working right now.
- Nothing that could start something in the house (a script, a scene, an automation) is accepted unless we can read what it does, and that's checked too.
- Locks, alarms, cameras, covers, valves, water heaters, and anything named for a garage, door or gate are refused by default. See what we won't touch.
What we keep
The read of your system and your report are stored encrypted and deleted automatically after 90 days. Your payment details go to Stripe and never touch our server. The full detail is in our privacy policy.
Questions
Can you turn my lights or appliances on and off?
No. The service changes configuration: automations, integrations, registry entries, the history database, backups. It doesn't operate your devices, and the Connect integration refuses those requests in your house before they reach Home Assistant.
Can I stop a job halfway?
Yes. Remove the Connect integration, or delete the token from your profile, and access ends immediately. You can also press Talk to a person on your job page at any time.
What if I don't approve anything?
Then nothing changes, and you still get the report. The Check-up is built to be used exactly that way.
Do you keep access after the job?
No. No access is held between jobs. A later job starts again with fresh access.
