The Llamatik app can act as a server for your other devices. Turn on Settings → Share this device and any other Llamatik app on the same network can send prompts to this device and use the models you already have loaded, the same way it would use a Llamatik Server. Use it to let a phone borrow your desktop’s GPU, or to share one big model between several devices.
Sharing is off by default and is never started automatically: you turn it on each time.
Supported platforms
| Platform | Hosting |
|---|---|
| Android | Yes. Runs in a foreground service with wake and Wi-Fi locks, so it keeps serving while the app is in the background. |
| Desktop (JVM) | Yes |
| iOS | Yes, foreground only. iOS suspends apps that leave the foreground, so the app keeps the screen awake while sharing. The host is advertised over Bonjour. |
| Web (Wasm) | No. A browser cannot open a listening socket. The option is hidden. |
What is shared
- Text generation and embeddings, using the models currently loaded in the app. Loading, unloading and reconfiguring models stays on the host; remote clients cannot call admin endpoints (they get
403). - If you are generating in the app yourself, remote requests get
503 Retry-Afterinstead of interleaving with your chat.
Other modalities (image, vision, audio, video) are present in the server library but are not exposed by the app host yet.
Turn it on
- Load a model in the app (the Models tab).
- Open Settings → Share this device and switch Share this device on. You can change the port (default
8080). - The screen shows the address other devices can connect to, the models being shared and the number of requests served.
Approving devices
Every device must be approved by you before it can use yours:
- On the other device, open Settings → Connect to Server and connect to this device (discover it on the network or enter its address; see Connecting to a server).
- This device shows an Allow this device? prompt with the other device’s name. Tap Allow or Deny.
- Approved devices are listed under Approved devices and persist across restarts. Tap Revoke to remove one.
If the request is not approved in time, the other device shows The request was not approved in time.
Heads-up: traffic between devices is not encrypted. Only approve devices you trust, on a network you trust. Generating for others also uses your battery and may heat the device.
Seeing who is connected
While sharing is on, the app shows the host status in two places:
- Settings shows a line under Share this device:
Sharing · no devices connectedSharing · 2 connectedSharing · 2 connected · generating for them
- Chat top bar shows a small share icon next to the server status dot:
| Icon | Meaning |
|---|---|
| Grey | Sharing, but nobody is connected |
| Green with a number | That many devices are connected |
| Pulsing orange | Generating for another device right now |
A device counts as connected if it made a request in the last 30 seconds. Connected clients check in about every 10 seconds, so a device that closes the app or loses the network drops out after roughly 30 seconds. The status refreshes every 5 seconds.
For developers
The embedded server is built from two modules in the Llamatik repository:
:server-lib(published ascom.llamatik:server): the engine-agnostic Ktor inference API shared with the standalone Llamatik Server. Entry point:installLlamatikServer(LlamatikServerConfig). It includes per-device approval (PairingManager,/v1/pairing/...).:host:EmbeddedServeron Ktor CIO, plus the Android foreground service and the iOS/desktop host controllers.
The server shares the app’s LlamaBridge singleton, so remote clients always use the models already loaded in the app.