Language models thatnever leave the phone.
SLM Store is a mobile app for browsing, downloading, and running small language models entirely on-device. Airplane mode is the demo: no network, no cloud calls, no data leaving the phone.
Labs experiment · built by Lakeline Data · not a client engagement
- On-device inference
- Offline-first
- Quantized models
- iOS · Android
One download. Then the network is optional.
AFTER THE FIRST DOWNLOAD, EVERY PROMPT, INDEX, AND ANSWER STAYS ON THE DEVICE.
- phi-4-miniREADY2.1 GB · Q4 · reasoning
- gemma-3nREADY1.6 GB · Q4 · multimodal
- qwen-3 4B▼ 64%
Private by physics, not by policy.
- 01
Field work. Inspections, healthcare visits, logistics — the places that need AI most are the places with the worst connectivity. Models run at full speed with zero bars.
- 02
Regulated data. When the prompt contains patient notes or client financials, "we don't store it" isn't as strong as "it never left the device."
- 03
Cost. Inference at the edge is free after download. For high-volume, low-complexity tasks, a 2 GB quantized model beats a metered API.
Built for the field.
Inspections, healthcare visits, remote logistics — the same two use cases keep working at zero bars, because the models live on the phone, not behind an API.
Real use cases, all on the device.
Explain any photo — on the phone
A camera roll full of site photos becomes structured data: the local model explains each photo and extracts the details — objects, damage, serial numbers, readings, text. Everything runs on the small offline model; nothing leaves the mobile.
Exact answers from documents on the device
Search inside any documents sitting on the mobile — PDFs, contracts, manuals, notes. The model reads them locally and returns exact answers with citations to the file, page, and section, so every answer can be checked.
Need AI where the network isn't?
We'll scope which of your workloads fit a small model — and prove it on a device in your hand.