Unsomnia

Privacy & trust

The most private thing on your phone, treated that way.

We’d rather tell you exactly what happens than promise something vague. Here it is in plain words. The privacy policy is the legal version.

Who touches your dream

Only after you agree. Until then, recordings stay on your phone and the app still works as a journal.

Transcription

Groq runs Whisper to turn your voice into text. If Groq is down, Cloudflare Workers AI does the same job as a backup.

Analysis

Anthropic’s Claude reads the transcript to write the title, summary and lens observations.

Storage

Your journal lives on Cloudflare, encrypted at rest, so it’s backed up and syncs between your phone and the web.

None of them may use your dreams to train AI models, and neither do we. Their processing is under agreements with no-training terms.

What you control

  • Consent you can take back. Withdraw it in Profile and new dreams stop being sent. Recording and your journal keep working.
  • Audio. Delete one recording or all of them. Transcripts and analysis stay.
  • Deleting. Delete a dream or your whole account. Deleted dreams wait 30 days in case you change your mind, then they’re gone.
  • Exporting. A single dream as text or PDF, your whole journal as a file, your recordings as a download.
  • Lenses. Turn off any interpretive lens and it goes completely silent, everywhere.

What we never do

  • Train AI on your dreams.
  • Sell, rent or share your data, or show you ads.
  • Put what your dreams say into analytics. Usage counts never include dream content, names or titles.
  • Read your dreams. Nobody at Unsomnia does.
  • Contact anyone about you, whatever a dream contains.

The web app

app.unsomnia.ai loads no third-party scripts, runs no analytics and sets one cookie: the one that keeps you signed in. Dream text never appears in a web address or a page title.

Where we’re honest about limits

Your dreams are analysed on our servers, not on your phone. We tried on-device transcription; at 3 AM it wasn’t accurate enough to be worth showing you. End-to-end encryption isn’t possible while a server-side model reads the transcript, so we don’t claim it.

What we can promise is narrow and true: encrypted at rest, never trained on, deletable by you, and processed by named companies under no-training terms.