How Steady handles your family's data

Plain English. No legalese. Written by Jack, who built this.

I'm Jack. I built Steady because the most useful conversations a parent can have shouldn't have to happen with a generic AI that has no idea who you are.

Here's the tension I can't pretend away: good coaching needs to know your kid — their name, their age, the thing that's been heavy this month. That kind of context only works if you trust where it goes. And most apps have not earned that. I know. I'm a parent too.

So here's exactly what we do with what you tell us, in the order it matters.

1. No training on family data. Ever.

Not yours, not your kid's. Not aggregated, not anonymized, not "to improve the model." We pay for AI from Anthropic, OpenAI, and Google through the Cloudflare AI Gateway with zero-retention enabled — your messages aren't kept by them, and they aren't used to train anything.

2. You can see, edit, and delete everything Steady remembers about you.

There's no hidden profile. The Family memory page renders Steady's notes about your family as plain-English bullets — edit any line, delete any line, or wipe the whole memory in one click. Honest note: that page lands in the first week of beta. Until it does, email me and I'll send you everything Steady has stored about you, by hand, same day.

3. Two-click cancel. Full data export. No retention modal.

Settings → Cancel. That's it. Download your full conversation history and family memory as JSON or PDF on the way out. We delete the rest within 30 days.

Where your data lives

Cloudflare D1 (a SQLite database), hosted in the US.
Sign-in sessions and short-lived login tokens in Cloudflare KV.
No third-party analytics see the contents of your messages, your kid's name, or anything Steady has noted about your family. If we turn on product analytics for the small test group, it's configured server-side to record event names and plain counters only — never message content.
Backups are encrypted at rest.

Who can read what you write

Nobody at Steady reads your messages by default. There's no team dashboard, no monitoring queue, no one scrolling through your evenings.

I'm alerted personally only when the crisis classifier fires. That's the one time a human at Steady sees a message. It exists so that a parent in a genuine crisis moment doesn't fall through the cracks. The alert shows me the family, the conversation, and the single message that tripped it — and nothing else.

Coaches see only the messages you hand them. When you bring a coach into a conversation, you pick which messages to share. They don't see your family memory, your other conversations, or anything you didn't choose to show them.

What Steady asks for, and when

Day one: your kid's first name and age, and one sentence about what's been heavy lately. That's the whole ask.

The deeper context — how you were raised, what you don't want to repeat, what's actually worked before — only gets added if and when you decide to add it. Usually that's after a few conversations have already been useful, not before.

Every piece of context Steady stores is something you wrote. Steady doesn't infer or invent details about your family.

About the AI providers

Steady routes to Anthropic's Claude first, then OpenAI and Google Gemini as backups, all through the Cloudflare AI Gateway with zero-retention turned on. The developer API tiers we use don't train on what customers send them.

We don't use the consumer ChatGPT, Claude.ai, or Gemini apps on your behalf — only the developer APIs, which carry different and stronger data terms.

The crisis path

Every message runs through a crisis classifier before it reaches the main model. If it fires, you get a hard-coded response with 988 (the Suicide & Crisis Lifeline) and 741741 (the Crisis Text Line) — not a reply the model wrote. I get a real-time alert. This path is the same for every parent, every time, and it can't be switched off.

Those phone numbers are written directly into the code. They can't be changed by accident, by a model, or by a bug.

What we don't do

The negative space matters as much as the promises:

We don't sell data. To anyone. Ever.
We don't share your messages with advertisers, insurers, schools, or employers — or with any third party, including law enforcement, except under a valid legal order. If we ever get one, we'll tell you, unless the law forbids us from saying so.
We don't run ads inside Steady.
We don't use dark patterns to keep you subscribed. Cancel is two clicks.

The honest caveats

Steady is new. You're reading this in 2026, and the company is small. If something feels off, email me — [email protected] — and I'll reply within 24 hours during the founding-families phase.

I'm not a clinician. Steady's clinical advisor reviews the crisis logic every month. If you're in a crisis, please call 988 or text 741741 — not Steady.

Trust your gut. If at any point this app feels like it's overreaching, leave. Take your data with you.

— Jack, founder of Steady. [email protected]