‹ BackHN Continuity

Thread

Show HN: Breadcrumb, record everything on your mac + context manager for AI

49 points · 9 comments · jv22222

Loading the complete thread in the background. This saved snapshot is available now. Refresh

  1. jsh42 · · focus · HN ↗
    I use this. My focus in in wringing the most ROI possible from my AI Interns. Breadcrumb is like upgrading to a better model and a better brain simultaneously.
  2. mrkn1 · · focus · HN ↗
    For recording everything you do (super sensitive), I don't see another way than offline.

    Two small OSS tools that might slot right into your capture pipeline:

    1. textsnap for turning screenshots/screen captures into plaintext (OCR, no GPU/cloud)

    2. yapsnap for local CPU transcription + diarization of meeting audio/video — both single-command, fully offline after the first model download, which sounds like it'd fit your "nothing leaves the machine" design.

    Also, for the local classification/naming step you mentioned (currently Gemma 3 4B via ollama) — gutsy is a much smaller 0.8B model purpose-built for classification/decision tasks a la Jev, might be worth a look if you want to cut that footprint down.

  3. Bjartr · · focus · HN ↗
    This is what I wanted Windows Recall to be
    1. jv22222 · · focus · HN ↗
      Would you like me to port it to windows?
    2. jv22222 · · focus · HN ↗
      The plan is to port it to windows and linux. The architecture is already there.

      I need to decide between building a UI editor for moments, memories and meetings OR the port to windows/linux work.

      What do you think I should work on first? (Asking everyone actually!)

      1. jv22222 · · focus · HN ↗
        I think it's going to be the UI to manage your data as that will bring higher value to existing users. I'll make that surface in HTML which will enable porting to be easier.
      2. takenotice · · focus · HN ↗
        I came to the comments looking for windows/linux alternatives. Looking forward to the ports!
  4. aidiveyt · · focus · HN ↗
    Do all 30+ tools load every request? Claude Code's 24 built-ins already cost ~85,500 chars of definitions before any MCP server.
    1. jv22222 · · focus · HN ↗
      While they do technically ride every request, they're cached in all good clients, so usually pretty cheap. Also Claude Code just loads names and defers the full definitions until it thinks it needs one.
  5. itsmeduncan · · focus · HN ↗

    [dead]

  6. edimaudo · · focus · HN ↗
    what is the security like? It is reminds me MSFT recall
    1. jv22222 · · focus · HN ↗

      [dead]

  7. paul_irolla · · focus · HN ↗

    [dead]

  8. hypnagogicjerk · · focus · HN ↗
    See also Screenpipe (YC S26)

    <a href="https:&#x2F;&#x2F;github.com&#x2F;screenpipe&#x2F;screenpipe" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;screenpipe&#x2F;screenpipe

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.