Generate chat replies and images with Naidan on another device
From Naidan on one device, you can request chat replies or image generation from models running in Naidan on another. Naidan RPC connects the two browsers; the device running the models chooses which functions it shares.
On each device, open Naidan. If first-run setup appears, close it with the × at the top right before following step 1.
The device providing a model must use the Hosted edition; Standalone can only use a peer’s inference. The public HTTPS relay in step 2 is blocked by Local-only and Curated. If either device uses one of those editions, switch it to regular Hosted before step 1. Settings on another site address do not carry over. Open regular Hosted (external site)
1. Turn on the connection feature in both browsers
In each browser, open Settings → Developer → Experimental Features and select Enable Naidan RPC. The Naidan RPC settings tab appears. This switch alone starts no connection or inference.
2. Bring the two browsers together
One person chooses a meeting code and tells the other. In Naidan RPC, select Connect to a new peer on both devices. Enter the same Piping Server origin and that code, then select Start pairing in each browser. You can also use your own HTTPS relay when the edition permits that connection; Naidan does not start one.
Piping Server origin to try in both browsers: https://ppng.io Public relays and self-host instructions from the Piping Server project (external site)
3. Compare the entire verification text
Compare every character shown on the two devices through a separate trusted channel. Only when both texts match, select The entire text matches on each device. The connection is then marked Connected and initially lasts This connection only. To select it later in Chat settings, choose Trust this peer next time and save this connection. Saving it does not enable automatic connection or shared functions.
4. Choose what the model device provides
On the device running the model, open its connection under What this connection lets the peer use on this device. Turn on Chat inference, Image generation, or both, then select Apply provided methods. Nothing is shared until you apply a choice. On the other device, check Available from peer to see what it may use; your own sharing switches do not control the other device.
5. Select the peer for chat
On the device where you chat, open Settings → Connection. Set API Provider to Naidan RPC, choose the saved connection, select a model from that peer under Default Model, then Save Changes. Existing and new chats that inherit the global connection can now request answers from that device. If the model list is empty, check that the serving browser has a local chat model and has granted chat inference.
Use the peer for images too
If the serving device grants image generation, open the image workspace on the caller. Under Components, choose Naidan RPC in Compute with, select the saved connection and list its image models. Choose a model on that device, then generate. The peer receives the prompt and any initial or reference images used for that request; its model files stay on the serving device.
What crosses the connection
The Piping Server relays encrypted traffic between the browsers. The verified peer receives the chat request, selected conversation context and any attached images, or the image prompt and selected image inputs, and sends results back. The relay is a separate service from the model-running browser. Naidan RPC does not download models or manage files on the peer.
Which browser can provide inference
The hosted edition can provide its local llama.cpp chat and image generation functions when their models and runtimes are available. The Standalone edition can use a peer but cannot offer its own inference through this feature. A saved connection does not reconnect on startup unless you separately enable Connect automatically on startup; you can disconnect it in the Naidan RPC tab.