Beacon

Press

A voice journal that stays quiet until it's sure.

Most AI journaling apps reply to every entry. Beacon listens, and only says something about you once the same thing has come up on three separate days. It's made by Adonay Merkonnen, one independent developer in the UK.

The short version

Beacon is a private voice journal for iPhone. You tap and talk for a minute; Beacon transcribes it, files it by day, and throws the recording away. After each entry it shows you the topics it picked up — nothing more. Every night it reads across your entries, and only when the same theme has come up on three separate days may it name it, once, in plain words. If the evidence isn't strong enough, it stays silent. There are no streaks, no feed, and no chatbot.

Key facts

  • Platform: iPhone. In private beta on TestFlight.
  • Price: free during the beta.
  • Made by: Adonay Merkonnen, an independent developer in the UK, who designed and built the app, the servers and the pattern engine alone. No investors.
  • Audio: transcribed and discarded. Beacon never stores recordings.
  • Where data lives: the EU — database in Ireland, servers in Dublin.
  • Deleting: one action in Settings permanently removes the account, every entry and every pattern.
  • Optional lock: Face ID can be required every time the app opens.

What Beacon deliberately doesn't do

  • Reply to every entry, or chat back.
  • Score your mood, or use streaks to keep you coming back.
  • Diagnose, or give advice.
  • Keep your voice recordings.
  • Speak before a theme has come up on three separate days — or speak at all when it isn't sure.

Sensitive entries

Beacon isn't a crisis service and doesn't claim to be. If an entry contains explicit signs of a risk of self-harm, harm to others, or acute crisis, Beacon shows the Samaritans number (116 123) straight under what was said, along with 988 for the US and findahelpline.com for everywhere else.

The technology

Audio is transcribed by Deepgram. Topics and pattern write-ups come from Anthropic's AI models, under terms where content isn't used for training. OpenAI embeddings are used to find entries about the same thing. The full detail is in the Privacy Policy.

Assets