Create your first LLM Surface

Store your API key as a secret

Select Secrets in the dashboard sidebar, then select New Secret. Fill in the fields:

Select Create Secret. Note the Secret ID, auto-derived from the name and shown live as you type it, even before you save; you will select it from the API Key Secret field in the next step.

Start a new surface

Under SURFACES in the sidebar, select LLM. On the LLM Surfaces page, select Add LLM Surface. The Create new LLM Surface dialog shows a list of starter templates; the list varies by appliance. Select Start blank.

The surface builder opens with the canvas already populated with the full request and response chain, Access Point, Prompt Guard, Judge, the LLM node, Jury, a second Prompt Guard, and Response, plus two Expert Witness slots, all connected. Every stage besides the Access Point and LLM nodes starts disabled; only those two need configuring for a working surface, and later guides cover turning the others on. (If you picked a starter template instead, it may pre-wire the provider and model for you; skip ahead to step 4 once you’ve confirmed step 3’s settings match what you need.)

The surface is given a default name automatically. To rename the surface itself, use Configure Surface Features, reached from a button on the Access Point’s own settings panel, not that node’s Name field, which only labels the node itself on the canvas.

Configure the LLM node

The LLM node defines which provider and model this surface calls. Select it in the canvas to open its settings panel.

Set Provider to OpenAI and Endpoint to Chat Completions, then pick any model from the Model list, for example the newest GPT model shown (gpt-5.5 at the time of writing; the catalogue is hot-reloadable, so the exact list on your appliance may differ). Set API Key Secret to the secret you created in step 1.

Configure the Access Point node

The Access Point node defines how incoming requests reach the surface. Select it in the canvas to open its settings panel.

The panel shows a Pipe Route banner followed by these fields:

The Pipe Route banner at the top of the panel shows the composed full URL: the listen address, path prefix, and custom path combined. Copy this URL. You will use it in step 6.

Save the surface

Select the save icon in the toolbar. The surface is saved and becomes active immediately; no restart is required.

Send a test request

Replace <YOUR_APPLIANCE_HOST> and <YOUR_SURFACE_ROUTE> with the values from the route banner in step 4:

curl -k -X POST "https://<YOUR_APPLIANCE_HOST><YOUR_SURFACE_ROUTE>/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [
      { "role": "user", "content": "Say hello from the Agent Stream quickstart." }
    ]
  }'

Replace gpt-5.5 with whichever model you actually selected in step 3, if different.

A successful response returns 200 OK with the OpenAI chat completion body, plus x-agent-stream-* response headers carrying the trace ID, provider, model, token counts, and cost for this call. See Observability → for the full header list.