Create an API key in Novex, then call the endpoint from your own server. Each key is permanently tied to the one model you chose. Do not put a secret key in browser JavaScript or a public repository.

POST https://novex-ai.pages.dev/api
Authorization: Bearer nvx_your_key
Content-Type: application/json

Choose the model when creating a key. The key remains bound to that model; sending a model field does not change it.

Use the chat key you created for Novex AI 2.5, Flash, or Pro. Send the conversation so the model has the context from earlier turns.

const endpoint = process.env.NOVEX_API_URL || "https://novex-ai.pages.dev/api";
const apiKey = process.env.NOVEX_API_KEY;
if (!apiKey) throw new Error("Set NOVEX_API_KEY on your server.");

const response = await fetch(endpoint, {
  method: "POST",
  headers: {
    Authorization: `Bearer ${apiKey}`,
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    messages: [
      { role: "user", content: "Explain photosynthesis in two sentences." }
    ]
  })
});
const data = await response.json();
if (!response.ok) throw new Error(data.error || `Novex API error: ${response.status}`);
console.log(data.prediction);

{
  "prediction": "Photosynthesis is the process plants use to...",
  "model": "Novex AI 2.5",
  "usage": null,
  "calls_total": 1
}

Chat is request/response; the API does not keep a conversation session or stream tokens. Store the returned answer in your app and send the updated messages array on the next request. Alternatively, send a single input string for a one-turn chat. The API accepts up to 20 messages per request.

Chat requests can include standard tools and tool_choice fields. Definitions are forwarded to the selected model; if it returns a tool call, the response includes tool_calls. Your application is responsible for running the tool and sending its result in the next request.

Send {"action":"provider_health"} with a valid key to check whether the key is recognized and see whether the Novex inference service is configured. This does not run a model inference.

The public Novex API currently supports text chat and image generation. Image understanding is available in the Novex chat page, but is not supported by API keys yet.

Create an image-generation API key, then send an image prompt using input. The key determines whether Novex Image Generation 2.5 or 2.5 Pro runs.

const response = await fetch(process.env.NOVEX_API_URL || "https://novex-ai.pages.dev/api", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.NOVEX_API_KEY}`,
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    input: "A watercolor lighthouse at sunset",
    width: 768,
    height: 768
  })
});
const data = await response.json();
if (!response.ok) throw new Error(data.error || `Novex API error: ${response.status}`);
// data.image_url is a PNG data URL.

Supported dimensions are 256–1024 px with 1:1, 16:9, or 9:16 aspect ratios. Defaults are 768 × 768 for standard generation and 1024 × 1024 for Pro; invalid dimensions fall back to the selected model's default. Free accounts can generate up to 4 standard images per UTC day. Pro accounts can generate up to 8 standard and 4 Pro images per UTC day. Limits reset at 00:00 UTC.

• 400: invalid JSON, missing input/messages, unsupported model data, or image understanding requested through an API key. Check the request body.
• 401: missing, invalid, revoked key, or sign-in required to create a key.
• 403: the selected Pro model requires an active Novex Pro plan.
• 413: request too large. Shorten the chat history or reduce the text payload; API-key requests do not accept images for understanding.
• 429: rate limit or daily free allowance reached. Back off; do not retry in a tight loop.
• 502/503: temporary inference failure or service configuration issue. Retry later with exponential backoff.

Successful chat responses include prediction and model; successful image responses include image_url and model. Check response.ok and parse the JSON error body on failures.