Connect your model in 5 minutes

Every deployment has a permanent API key (sk-sm-…) and an OpenAI-compatible endpoint. Anything that can talk to the OpenAI API — official SDKs, LangChain, n8n, off-the-shelf bots — works with your model by just swapping base_url.

The key lives on the Deployments page → the </> button → "🔑 key". Keep it on your server, never in page code.

Python (openai SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://sculptmind.com/api/ext/v1",
    api_key="sk-sm-...",          # permanent deployment key
)

resp = client.chat.completions.create(
    model="my-model",             # any name: the key already picked the model
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

Streaming

stream = client.chat.completions.create(
    model="my-model",
    messages=[{"role": "user", "content": "Tell me about your company"}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta if chunk.choices else None
    if delta and delta.content:
        print(delta.content, end="", flush=True)

Reasoning models additionally send a delta.reasoning field in the chunks — the model's thoughts before the answer. Whether to show them to the user is up to you.

JavaScript / fetch

const r = await fetch("https://sculptmind.com/api/ext/v1/chat/completions", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "Authorization": "Bearer sk-sm-...",   // only from your backend!
  },
  body: JSON.stringify({
    model: "my-model",
    messages: [{ role: "user", content: "Hello!" }],
  }),
});
const data = await r.json();
console.log(data.choices[0].message.content);

What gets injected automatically

• Built-in system prompt of the deployment — set in chat with the "📌 embed" button, applies to every request.
• Knowledge base (RAG) — if documents are uploaded, relevant fragments are injected into every question. To disable: extra_body={"use_rag": false}.

Limits and behavior

• 60 requests/min per IP; responses up to 8192 tokens.
• A sleeping model responds with 503 model_not_ready — start it from the dashboard (auto-stop protects the owner's GPU hours).
• The key lives forever and does not depend on public links; rotate it in the </> panel with the 🔄 button (the old key dies instantly).
• GET /api/ext/v1/models — connection check.

No code

• Website widget — a one-line iframe (the </> panel on the deployments page).
• Public link — chat without sign-up, optionally password-protected.