DigUp for Journalists

Download DigUp for Mac EmbeddingGemma 2 All roles

On-device flow Schematic: DigUp uses EmbeddingGemma 2 for Journalists local search

Original schematic — DigUp indexes local files with EmbeddingGemma 2 (llama.cpp).

Sourced snapshot (2026-10-15)

Fact	Value	Source
Model	EmbeddingGemma 2 · ~740M params · ~865 MB Q8 via llama.cpp	Google AI · DigUp README
Vector space	768-d shared multimodal space	Google AI · DigUp README
Mac / Win·Linux	0.6.1 · 0.7.0-beta.1	GitHub Releases
Stars / forks	~3,768 / ~254	GitHub API

Pain → DigUp workflow → tip

Pain	DigUp + EmbeddingGemma 2 workflow	Tip
Where did they mention the budget?	Audio 30s windows	Jump near the quote
B-roll of the protest sign	Video frame embeddings	Find by what’s on screen
FOIA PDF needle	PDF page meaning + keywords	Open at the page
Source privacy	Local-first DigUp	No DigUp account; choose folders carefully

Numbered checklist

Install DigUp; index Interviews / Research locally (download cloud proxies first).
Finish ~865 MB EmbeddingGemma 2 download.
Query spoken topics; verify quotes in the player before publishing.
Use overnight indexing for multi-hour media.
See FAQ for model size / Use Files if network blocked.

Why EmbeddingGemma 2 matters here

Google’s EmbeddingGemma 2 (~740M, unified 768-d multimodal space) is what DigUp runs via llama.cpp (~865 MB Q8 package). One query can rank images, PDF pages, audio windows, video frames, and optional code — see the local multimodal guide and model page. digup.run is not Google-official.