Offline AI chat that actually works
The short answer
ShadowTalk AI is a free, installable web app where the AI model runs on your own device. After a one-time download (300MB-2.6GB depending on your device tier), chat works with zero internet — and anything you send while offline queues up in durable local storage and syncs the moment you're back. You install it in one tap from the browser: no app store, no account, no card.
Most AI assistants die the moment your signal does. ShadowTalk is built the other way round: the model lives on your device, the app shell is cached, and the network is only needed for the very first visit and for optional cloud work. This guide explains what works offline, what still needs a connection, and why the offline-plus-sync combination is something the rest of the market doesn't ship.
Can you use ChatGPT offline?
No — ChatGPT, Gemini and Claude are cloud services: every token is generated on their servers, so without a connection they simply cannot answer. That's true even for their mobile apps with history cached.
The offline options that do exist each give up something. Desktop apps like LM Studio and Ollama run powerful local models but keep chats on one machine — LM Studio's device sync is still an open feature request (issue #2467) and ChatterUI's WebDAV chat-merge never shipped. Mobile apps like PocketPal or MuraAI run models on your phone but don't sync either. Browser demos like Secret Llama or web-llm-chat run locally with no accounts — but nothing backs up or moves your history.
ShadowTalk is the first AI chat in a browser that works offline and syncs when you're back: on-device WebGPU inference for the answers, durable local caching for the chats, and automatic catch-up sync the next time you have signal — with end-to-end encryption while synced when Private Mode is on.
Built for real dead zones
Tunnels, flights, rural coverage gaps, blackout zones, campus firewalls — ShadowTalk assumes the network is a sometimes-food. Because inference runs on your GPU through a background Web Worker, replies stream at the speed of your hardware, not your signal bar.
- Reloads lose nothing: chats and drafts persist in durable on-device storage (verified by automated offline smoke tests in CI)
- The app shell loads with zero third-party requests — fonts, scripts and even model weights come from ShadowTalk's own infrastructure
- Interrupted model downloads pause and resume shard-by-shard; a dying connection never restarts from zero
- Chats sent with no signal queue locally and arrive when the network returns — no re-sending, no lost drafts
The part everyone else is missing: sync
Offline AI and synced AI chat have never shipped in the same product — until now. The landscape as of November 2026, each line checkable:
- Offline AI camp (LM Studio, web-llm-chat, Secret Llama): real on-device models, no sync at all — LM Studio device sync is feature request #2467, ChatterUI's WebDAV merge is unshipped
- Encrypted AI camp (Proton Lumo, Confer): synced encrypted history — but cloud-only inference, an account, and a live internet requirement; Lumo's own security model documents the server decrypting your prompt in cleartext to answer it
- Local + sync attempts (Locra and Pal Chat via iCloud, LocalMind via your own Ollama/LM Studio server plus your own S3): native apps or bring-your-own-server — never just a browser tab
What you need (honest requirements)
On-device chat needs a WebGPU-capable browser — Chrome or Edge 113+ — and enough storage for the model tier your device picks. The very first visit needs a connection to install the app shell and download the model once; everything after that runs locally. Cloud models and real-time web search always use the network, and devices without WebGPU fall back to cloud chat when online.
- Browser: Chrome or Edge 113+ (WebGPU enabled by default)
- Storage: about 600MB-2.6GB by model tier — the manager pre-checks your quota before downloading
- Account: none required — guests get a local model automatically
- iOS note: the PWA installs, but WebGPU support on iPhones varies by iOS version; on-device mode targets Chrome/Edge on desktop and Android first
Start chatting offline in about a minute
- 1Open shadowtalk-ai.com in Chrome or Edge 113+ while you still have a connection.
- 2Install the app: use the install icon in the address bar (PWA) — it lands on your home screen or dock.
- 3Open the model manager, pick the model your device suggests, and let it download once — interrupted downloads resume automatically.
- 4Go offline. Keep chatting: replies are generated on your device with zero network requests.
- 5When signal returns, queued messages sync on their own — end-to-end encrypted if Private Mode is on.
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.
Start offline AI chat — free