digup: 865MB local semantic search over files with EmbeddingGemma 2, fully offline
virtualunc · reddit · 2026-10-22
A Redditor surfaced digup, a 12-day-old MIT-licensed local semantic file search tool already at 5.3k stars, built on EmbeddingGemma 2 (740M params, Apache 2.0).
Highlights:
- 865MB one-time model download, then fully offline — no account, no telemetry (daily update check can be disabled)
- Text, images, audio, and video share one embedding space: a photo of a dog on a beach is findable by words; 100+ languages both ways, so an English query finds Bengali notes
- Exact string matching coexists with semantic search (invoice numbers, error codes); on Mac it even reads text inside screenshots
- Results open in place: PDFs at the page, audio/video at the timestamp, code at the line
- macOS dmg/brew (Apple Silicon, macOS 14+), unsigned Windows beta installer, Linux deb/appimage
Unknown: index size on disk and RAM usage during indexing — the author asks big-library users to share numbers.
More from Apps
- Zuck says shopping is Muse AI's top use case: 25%+ of users buy through the agent, averaging $200+ — Polymarket · 2026-10-22
- Lila Sciences showcased in NVIDIA GTC Berlin keynote, targeting real-world verification for scientific superintelligence — tw_killian · 2026-10-22
- Datalab launches word confidence model to verify OCR accuracy on messy scans — VikParuchuri · 2026-10-22
- Meta says Muse tops 10M users, with shopping a leading activity — TechCrunch AI · 2026-10-22
- Four cloud agents in two months: how Claude Duet, Grok Bot, Muse, and Dots actually differ — dotey · 2026-10-22
- wabi demos agent that auto-builds trip itinerary apps, polls groups and books plans — joonasvirtanen · 2026-10-22