# local semantic file search for Linux (Rust, llama.cpp, EmbeddingGemma 2) — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/50d08826a2fa76b1

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-11 01:52 ingested 2026-10-12 02:39

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1x2vz3y/local_semantic_file_search_for_linux_rust/)

## retained text



```
indexes PDFs, Office files, scans, screenshots, audio and video, and finds them by content. A query like "when did they decide to freeze hiring" opens the recording at that second.

\* EmbeddingGemma 2 embeds text, images, video frames and 30 s audio windows into one vector space, run in-process via llama.cpp (CUDA, Vulkan or CPU). Ranking is hybrid: vector similarity plus SQLite FTS5 for exact strings. OCR is PaddleOCR. Engine in Rust, app in GTK 4. No LLM, no network after the model download.

On an i5 laptop with Iris Xe: right file first on 95% of a small test set, \~4 ms keyword results, \~0.17 s semantic search, Vulkan indexing \~3x faster than CPU.

Missing: audio transcription, packages, Wayland shortcut on older desktops.

MIT: [https://github.com/pdstn12/Recall](https://github.com/pdstn12/Recall)

Inspired by \[DigUp\](https://github.com/ARahim3/DigUp), a macOS app by Abdur Rahim.

https://i.redd.it/7jvvwjndvquh1.gif
```



document fetched 2026-10-12 00:37 · canonical https://reddit.com/r/LocalLLaMA/comments/1x2vz3y/local_semantic_file_search_for_linux_rust
