Quickstart¶

This path connects one OpenAI-compatible server and sends a first message.

1. Start Diogenes¶

./startwithuv.sh

Open http://localhost:7000, complete the first-user form, and sign in.

2. Register a model endpoint¶

Open Settings → Models → Add endpoint and enter a base URL such as:

http://127.0.0.1:8000/v1

The endpoint form maps to POST /api/model-endpoints in routes/model_routes.py. A useful local entry has:

Diogenes normalizes the base URL, probes the model list, and stores the endpoint. A bare local base such as http://127.0.0.1:8000 is translated to /v1/models for discovery and /v1/chat/completions for chat.

The same registration can be performed from an authenticated browser session:

const body = new FormData();
body.set("name", "local-vllm");
body.set("base_url", "http://127.0.0.1:8000/v1");
body.set("model_type", "llm");
body.set("endpoint_kind", "auto");

const response = await fetch("/api/model-endpoints", {
  method: "POST",
  credentials: "same-origin",
  body
});

console.log(await response.json());

3. Create a chat¶

Use the model picker in the sidebar. GET /api/models returns models visible to the signed-in user, removes hidden entries, and preserves pinned entries for the picker.

Create a session, type a short prompt, and submit. Normal chat uses POST /api/chat_stream; the browser reads server-sent events and stores the completed transcript in SQLite.

4. Check the route¶

The model label shown under the answer is derived from response metadata. When a configured foreground fallback answers, the message records both requested and responding endpoint/model values.

Endpoint errors¶

Use Settings → Models → Test or GET /api/model-endpoints/{id}/probe before debugging chat rendering.