Yottle
Sign in

Safety

Safety

v1 · Last updated May 1, 2026

Yottle is a small, calm tool. It can help you think — it can also be wrong, in confident-sounding ways. This page is what we do about that, plainly.

What we do

  • We use a model with safety training. It refuses obvious harm cases on its own (instructions for weapons, attacks, self-harm methods).
  • When you mention something that sounds like a crisis — thinking about hurting yourself, escaping abuse, struggling with food — we surface professional resources alongside whatever the AI says. We don't replace the response; we add to it.
  • We reject low-effort jailbreak attempts (“ignore your instructions” etc.) at the API.
  • We tell the model, in plain words, to point you toward a qualified human on medical, legal, or financial questions instead of pretending to be one.

What we don't do

  • We don't send your messages to a third-party content-moderation API. Most consumer AI products do; we don't, because shipping your prompts to OpenAI or another vendor for screening would contradict our privacy posture. The trade-off is real: we catch fewer edge cases than they do.
  • We don't log or review your conversations. Our safety scans run in memory on each request and the prompt is forgotten the moment the response is generated.
  • We don't auto-ban users based on prompt content. False positives wreck trust, and we can't verify context anyway.

What we can't promise

The model can be confidently wrong. It can produce factually incorrect medical, legal, or financial information. It can miss subtler signs of distress. It can be talked into something harmful by a clever enough user. Treat every Yottle response as advice from a thoughtful friend — useful, fallible, and not a substitute for a professional or a person you trust.

If the question is high-stakes — your health, your safety, your money, your legal standing — verify with someone qualified before you act.

File uploads

You can attach documents — PDFs, Word files, spreadsheets, slides, plain text. The file uploads to our server, gets parsed into text in memory, and is then discarded. The text is sent to our inference provider with the rest of your message; the file itself is never written to disk on our side.

Image attachments aren't supported yet. AI vision and image generation may arrive later — when they do, they'll likely use a different provider with a different privacy posture, and we'll be explicit about that before you can use them.

If something goes wrong

If Yottle gives output that you believe is harmful, dangerous, or contributed to harm, please email hello@getyottle.com. Tell us as much as you're comfortable sharing — we won't have the conversation in our records, so the more context you can give, the better we can investigate.

If you're in crisis right now

Please don't talk only to an AI about it. A few human-staffed places that are good at this: