AI features

What runs on-device

Three things: smart-categorize (the model proposes a kind and pre-fills fields from a scanned payload), read-ticket (vision-capable models extract structured fields from a photo), and semantic duplicate detection (embeddings find stashes you may have already saved).

Install the embedding model

AI Setup → Install Embedding. The Gemma 300m embedding model is about 175 MB. It powers semantic duplicate detection and the semantic search bar on the home screen.

Pick a generate model

AI Setup → Generate. Three options: Gemma 3 1B (≈500 MB, the default — works on every supported device), Gemma 4 E2B (≈2.5 GB, faster on capable phones), and Gemma 4 E4B (≈3.5 GB, the largest).

iOS Gemma 4 limitation

Gemma 4 E2B and E4B currently crash on iOS because of an upstream flutter_gemma bug with the .litertlm bundle format. Snikko hides Gemma 4 on iOS until the upstream fix ships — only Gemma 3 1B is selectable there.

Auto-categorize on scan

With AI features enabled, scanning a code routes the payload through your installed generate model. The model proposes a kind (Parking / Transit / Loyalty / Event / Wi-Fi) and pre-fills the matching fields. You can accept, edit, or override before saving.

Smart-create from a description

On the Add screen, tap Smart Create and type a free-text description ("Concert ticket, BLACKPINK Bangkok, Section A2, Row 14, March 8"). The model returns a structured stash that you can review and save.

Disable AI globally

We can turn the AI feature set off remotely through Firebase Remote Config. When AI is disabled, the AI badges, Smart Create entry point, and Read-ticket option are hidden — the rest of Snikko works exactly as usual.

Delete a model to free space

AI Setup → tap the installed model → Delete. The file is removed from app storage immediately. You can re-download it any time from the same screen.