Show HN: I made a Mac app to search my images and videos locally with ML

Original Article ↗ Hacker News Discussion ↗

A new macOS app, Desktop Docs, lets users semantically search local images and videos using CLIP-based machine learning models that run entirely on their own machines. Commenters praise its offline, one-time-purchase model and potential for cross‑platform support, while raising questions about performance, index size, UI design, and overlap with tools like Spotlight, Apple Photos, and Immich. Many request features such as PDF and text search, facial recognition, NAS support, and duplicate detection, and several insist on a trial version before paying.

App concept & core capabilities

  • Desktop app to semantically search local images and videos using ML embeddings (CLIP).
  • Works fully offline; developer states media never leaves the machine.
  • Videos are sampled into frames (about 1 frame/sec), resized (e.g., 256×256), embedded into 512‑dim float32 vectors for search.
  • Current focus is on semantic/visual search rather than exact text or filename search.

Local ML, privacy, and performance

  • Multiple comments appreciate that it is 100% local, one‑time purchase software.
  • Some worry about model size and resulting large app footprint.
  • Indexing reportedly takes “a few minutes” depending on library size; optimization is ongoing.
  • Users request very explicit guarantees that nothing is sent over the network.

Platform, tech stack, and “Mac app” debate

  • App is built with Electron. This sparks a heated debate:
    • Some argue Electron apps are still “Mac apps” (they run on macOS, easy cross‑platform).
    • Others say Electron apps feel alien on macOS and don’t meet Apple’s HIG expectations.
  • Cross‑platform potential is viewed positively; Windows and Linux versions are in progress.

Overlap with existing tools

  • Comparisons to Spotlight, Photos, Screenie, Immich, and other tools.
  • Spotlight in newer macOS versions has some visual and OCR search, but:
    • It’s seen as limited in “concept”/semantic search and inconsistent in practice.
    • Desktop Docs is perceived as more focused on rich semantic media search.
  • Apple Photos already does strong people recognition, but some don’t want iCloud or Photos lock‑in.

Requested features / roadmap

  • Popular asks:
    • Text/PDF indexing, semantic text search.
    • Facial recognition and user‑taggable people (e.g., “grandma and uncle together”).
    • Near‑duplicate detection and smarter photo culling (blur, eyes closed).
    • Metadata‑based queries (camera model, resolution, SD/HD/4K) and richer tagging.
    • NAS support, Umbrel/Nextcloud/ownCloud integration.
  • Developer indicates many of these are “on the roadmap” and may be prioritized by interest.

Pricing, trials, and refunds

  • One‑time price currently around $24.99; previously higher prices caused confusion.
  • License includes future upgrades.
  • No trial version; several users say a time‑limited or image‑limited trial is essential.
  • Non‑refundable policy draws concern; others note chargebacks and goodwill as practical constraints.

UI/UX and marketing feedback

  • Suggestions:
    • Emphasize “local ML” rather than generic “AI” to build trust.
    • Clearer landing page with more screenshots/video demos.
    • Persistent “Buy” button instead of scattering CTAs.
  • Overall sentiment: interesting, useful idea with strong enthusiasm, tempered by concerns about Electron, overlap with Apple’s tools, lack of trial, and feature gaps.