Connect to an OpenAI-compatible server

Use this route when an inference server or API exposes OpenAI-compatible model and chat endpoints. Start with a server that is already running and has a model available. Naidan connects from your browser; it does not start the server. When you send a chat message, the message and the conversation context needed for the reply go to the server at the URL you chose.

First, open Naidan. If first-run setup appears, close it with the × at the top right, then follow step 1.

The local presets fill in LM Studio at http://localhost:1234/v1 or llama-server at http://localhost:8080/v1. Choose the preset that matches your server, or enter its actual base URL. A remote API may require custom HTTP headers. Its authentication, CORS policy and model support depend on that service.

1. Open Connection settings

Select Settings at the bottom of the sidebar. The Settings panel opens on Connection by default; choose Connection if another tab is active.

The sidebar Settings button opens the Connection panel.

2. Choose the provider and URL

Set API Provider to OpenAI Compatible. Choose the LM Studio (local) or llama-server (local) preset if it matches your server. Each preset sets the provider and URL together. For another compatible service, type its base URL, including a path such as /v1 when that service requires it. Local-only and Curated block arbitrary external API URLs; use an edition such as regular Hosted that permits the chosen service.

A local preset fills the base URL. The example shows LM Studio; use your server’s actual address.

3. Add headers only when your service needs them

Under Custom HTTP Headers, select Add Header and enter the name and value required by your service. Naidan sends these headers with its requests. Skip this step if your server does not require headers. Keep credentials out of URLs you share.

Add Header reveals name and value fields. This local example removes the empty row before checking.

4. Check the connection

Select Check Connection. Naidan requests the server’s model list and briefly displays Connected after a successful response. In Settings, changing a localhost address can also start this check automatically. A failed check may indicate a wrong URL, an unavailable server, or a browser CORS, origin or CSP restriction.

The model-list request moves from checking to Connected on success.

5. Choose a default model and save

In Model Selection, choose a model returned by this server under Default Model, then select Save Changes. The example model label stands for a result from your server. The default applies to existing and new conversations that inherit it; a chat or group can override the model.

Choose a returned model and save the Connection settings.

If the model list does not appear

Check the server’s base URL and its own model list first. Naidan appends /models for the check and /chat/completions for chat. If the page runs from file://, the browser can send Origin: null; a hosted page can have a Content Security Policy limiting destinations. The server must accept the browser’s request and any headers it requires.