2
5 Comments

I started voice-journaling because I couldn't sleep. Then I rebuilt the app around it to run entirely on-device.

I moved to a different country earlier this year for my studies. Somewhere around then I started having nights where my head wouldn't stop, one
thought leading to the next, lying there for hours. Not every night. Maybe
once every couple of weeks at first, then more often.

I'd never journaled in my life. Tried typing it out at a laptop for a few
days — lasted three, because I kept editing myself into the version of the
day I wanted to remember, not the one I actually had. Switched to just
talking into my phone before bed instead. That one stuck. Two and a half
months in, I've made 37 entries across 75 days. Skipped almost half the
days and it didn't matter — voice doesn't let me self-edit the way typing
did, and that's the whole reason it works.

I built Weave to make that habit easier to keep. First version was a cloud
AI web app — worked fine, until I actually sat with what that meant: every
one of those 2am entries was going out to a model provider I didn't own.
That's not a small thing to hand over. So I've spent the last few weeks
rebuilding the entire reflection pipeline to run on the phone itself.

Stack: Llama 3.2 3B (Q4_K_M) via llama.rn, Whisper small.en via whisper.rn
for transcription, MiniLM for semantic embeddings — all local, SQLite,
nothing synced anywhere. Metal on iOS, CPU on Android. Works fully in
airplane mode after the one-time model download.

Numbers from device testing: iPhone 17 goes save-tap to a complete
reflection in ~11.6s. OnePlus 9R on CPU is ~59s — still working on that
gap. Reflection quality on a 3B model isn't going to out-nuance GPT-4, and
older phones sit right at the memory edge. Wanted to say that upfront
rather than have someone find it themselves.

iOS TestFlight is open: testflight.apple.com/join/wpPKnRXg — if the "talk
for two minutes before bed" habit sounds like yours, I'd love for you to
try it and tell me what's broken.

Website : https://www.tryweave.app/
iOS Testflight: https://www.tryweave.app/beta

on August 6, 2026
  1. 2

    Moving the entire reflection pipeline on-device is a pretty consequential product choice.

    Have early testers shown you yet whether privacy is actually one of the reasons they choose Weave, or is that still something you’re waiting to learn from the beta?

    1. 1

      Honest answer: no hard data yet, this is still early beta. What I do have
      is anecdotal — a handful of people who tried it and mentioned privacy
      unprompted when I asked what they thought, plus my own experience
      building the first version.

      That first version had cloud AI, and the thing that got me was realizing
      I didn't want my own entries leaving my phone — I journal for the first
      time in my life with this app, and something about a private thought
      sitting on a server I don't control, even one I trust, felt wrong once I
      actually thought about it. That's really what pushed the rebuild, not a
      market research finding.

      So no, I can't tell you yet whether it's a real acquisition driver or
      just something I personally care about. Hoping the beta gives me an
      actual read on that, too early to say with confidence either way.

      1. 1

        That’s helpful context. The distinction between a principle you personally care about and a value users actively choose the product for is an interesting one to validate.

        I’d like to continue the conversation outside the thread. What’s the best email to reach you on?

        1. 1

          adityasanap182107@gmail.com - you can contact me using this email.

          1. 1

            Thanks! I’ve just sent it over.

            Looking forward to hearing your thoughts whenever you have a chance.