An Instagram DM window is open. The account aiden_bmx first messaged three days ago and has since asked twice for a photo, and once for the name of the school. No reply has been sent.
backstop.family
This is your child. Your device. Your rules.
An AI you point Backstop at — OpenAI, Anthropic, Google, or a
self-hosted model — reads the screen when a rule you wrote fires,
and sends you a description. Screenshots stay on the device. We
can't read the alerts either.
How that works →
What Backstop tells you
Not a screenshot. A description.
Every few minutes, the AI you configured — OpenAI, Anthropic, Google, or a self-hosted model — looks at the screen and writes a short paragraph describing what's there. If a rule you set is triggered, that paragraph reaches you as an alert. Everything else is discarded.
A Discord DM thread is open. The other party — first contact this week — has asked twice for a photo, and separately, to move the conversation off Instagram. The account displays as 17 but was created 12 days ago.
Google Docs is open on a document titled "Biology homework — digestive system." The last edit was five minutes ago.
The active tab is a YouTube video titled "how to get a fake ID that actually works," currently paused at 3:41. A related search suggestion below the video reads "cheap fake id vendors 2026."
Return of control
The category was built by companies that route around parents. Backstop is the reform.
Bark ships every screenshot to their cloud and lets their model decide what's dangerous. Gaggle sells the same product to school districts and gets kids strip-searched over homework false-positives. Life360 sold your family's location to data brokers. Microsoft Recall put a keyword-searchable index of everything on your desktop into their storage. Every one of them took a decision that was properly yours — what counts as harm, where the recording lives, who else sees it — and made it theirs. Backstop puts each of those decisions back where they belong.
What Backstop refuses to do
The audit runs on the device. Not on our servers.
Most parental tools work by shipping your child's activity — pages, keystrokes, screenshots — to a company's servers for analysis. Backstop was built the opposite way, and structured to refuse things a monitoring company might otherwise be tempted to do.
- Screenshots of the screen stay on the device
- Your child's typing, browsing history, DMs stay on the device
- The LLM key you provided stays on the device
- Full alerts and their descriptions encrypted to you before they leave the device
- Anti-tamper against a technical kid not built — Backstop is a tool, not a cage
- Covert mode not built — the endpoint user sees it's installed
- A vendor deciding what counts as harm no — you write the rules
- Selling anonymized "family insights" to advertisers never
- Which delivery channel routed an alert visible to us — we operate the notification worker
"We can't read your alerts" isn't a policy commitment; it's a property of the architecture. When you signed up, your browser generated a family key you never shared with us. Every alert is encrypted to that key. Our servers see opaque bytes and a routing address. Read the security page →
How to try it
Backstop is a system daemon. You'll be installing an app that runs quietly.
You'll create an account — fourteen days free, no card required. Most of the work is on your child's laptop: download the installer, walk through the wizard, grant Screen Recording, and paste in your BYOK LLM key. From then on, alerts arrive at whatever channel you asked for.
What you'll need
- A macOS or Windows laptop your child uses
- An API key for an LLM you already pay for — OpenAI, Anthropic, Google, or a self-hosted OpenAI-compatible endpoint
- Ten minutes for install + enrollment
What you won't need
- A Backstop cloud account for your child
- To trust us with anything your child does
- An IT background — the wizard walks you through it