Offline AI on your phone — no app store, no sync gap
The short answer
ShadowTalk AI runs offline on iOS and Android as an installable PWA: one tap from the browser, no app store, and the model runs on your phone's GPU with zero network egress after a one-time download. Chats you send offline queue up and sync when you're back — and because the workspace is the same web app on every device, phone and desktop history catch up with each other instead of living on separate machines.
Offline AI on mobile usually means a native app from an app store — and native on-device apps share one structural flaw: their history is trapped on the device. This page covers what offline AI actually looks like on a phone in ShadowTalk, what your hardware can handle, and the limits we'd rather tell you now than after you install.
Native offline AI apps have a sync problem on phones
Mobile on-device apps like PocketPal run GGUF models on your phone — real offline AI, and we credit them for it. But their chats live on that one phone: nothing flows to a desktop, nothing backs up, and switching devices means starting over. That's the same gap desktop tools like LM Studio have (device sync is still open feature request #2467 there), repeated on the device you actually carry.
ShadowTalk approaches it from the other side: the phone gets the same web workspace as every other device, the model runs on-device, and the durable offline queue plus catch-up sync means what you do on the train arrives on your desktop.
What offline looks like on a phone
- Install is one tap: open the site in a mobile browser and add it to your home screen as a PWA — no store, no review lag, no separate download
- The app picks a model sized to your device tier — roughly 300MB to 2.6GB, downloaded once over Wi-Fi in resumable shards
- Inference runs in a background worker on your phone's GPU — replies stream at hardware speed, not signal speed
- Messages sent with no signal queue in durable local storage and sync when the network returns
- Reloads, force-quits and restarts lose nothing — the cache is built to survive, verified by automated offline smoke tests
Honest limits on mobile hardware
Phones are not GPUs with screens, and pretending otherwise is how on-device AI gets a bad name. The practical ceiling for smooth on-device chat is the 1B-3B class — strong for drafting, summarizing, rewriting and Q&A, not frontier reasoning. WebGPU on mobile means a current Chrome or Edge on Android (113+); iOS Safari can run the PWA and its cloud tiers, with on-device support dependent on Apple's WebGPU rollout as of November 2026. Long sessions warm the phone and spend battery like any 3D game would — the GPU is doing real work. And the first visit plus model download need a connection; offline starts on the second.
Install offline AI on your phone
- 1Open shadowtalk-ai.com in Chrome (Android) or Safari (iOS)
- 2Add to Home Screen — Android: menu → 'Install app'; iOS: Share → 'Add to Home Screen'
- 3Open the installed app and pick the on-device model; let the device-tiered model download over Wi-Fi once (shards resume if interrupted)
- 4Chat — from here on, no signal required; offline messages queue and sync when you're back
Frequently asked questions
Keep reading
Free to use — no account, no card.
Works offline. Syncs when you're back. Your keys, your models, your chats.
Install on your phone — free