About and method

SiliconShelf is a prototype of an independent directory for apps that run AI models on Apple devices. It is run by one person in Austria and has no audience data to show yet.

Where the data comes from

Current snapshot (2026-10-16): 412 apps, of which 14 hand-picked seeds and 398 found by keyword search; 17 have a linked public repository; 40 are flagged stale. 1062 raw search results were considered; 651 were dropped by the relevance filter.

How tags work

Tags are detected by keyword matching on listing text (English and German) and then discarded; we do not store or republish store descriptions. A tag means "the listing mentions this", not "we tested this". Typical failure modes: an app supports Shortcuts but never says so, or mentions Siri only to say it is unsupported.

On-device
Listing text suggests models run locally on the device.
Offline
Listing text mentions working offline / without internet.
GGUF
Listing text mentions GGUF model files.
MLX
Listing text mentions Apple's MLX framework.
llama.cpp
Listing text mentions llama.cpp.
Apple Foundation Models
Listing text mentions Apple's Foundation Models framework.
Shortcuts
Listing text mentions Shortcuts / App Intents.
Siri
Listing text mentions Siri.
Server client
Listing mentions Ollama or a similar server; models may run on another machine.
Open source
A public GitHub repository is linked to this app (see the repo for the license).
Open source (claimed)
The listing says open source, but no repository is matched yet.

Stale, price and ratings

Summaries

Each summary is generated from structured fields (developer, platforms, price, dates, detected tags). It is not the developer's text and not a review.

Independence

No listing is paid for today. If sponsored placements are ever offered they will be labelled and will not change tags, stale flags or alternatives. App Store links are plain links, not affiliate links.