Your first LLM API request

Start with the route contract

Use the API key dashboard to obtain a server-side key with access to your intended model. Keep it out of client bundles, screenshots and source control. Set API_KEY in your shell before running these examples. Discovery is read-only; generation requests incur usage.

curl --fail-with-body "https://newrouters.com/v1/models" \
  -H "Authorization: Bearer $API_KEY"

Use /v1/models to find an exact public model ID and its declared protocol and capabilities. This discovery response does not list supported endpoints. For this example, select a model with protocol: "openai", then confirm in its current model guide that it supports POST /v1/chat/completions. The protocol label alone does not guarantee endpoint compatibility, and catalog visibility does not guarantee that your key can call the model. Check account access before continuing.

Send a minimal request

Save this as first-request.mjs and set MODEL_ID to the verified public ID. This example uses Node.js built-in fetch and performs one request without automatic retries.

const { API_KEY, MODEL_ID } = process.env;
if (!API_KEY || !MODEL_ID) throw new Error('Set API_KEY and MODEL_ID');
const response = await fetch('https://newrouters.com/v1/chat/completions', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: MODEL_ID,
    messages: [{ role: 'user', content: 'Explain HTTP in three short steps.' }],
    stream: false,
  }),
  signal: AbortSignal.timeout(60000),
});
if (!response.ok) throw new Error(`Request failed: HTTP ${response.status}`);
const result = await response.json();
console.log(result.choices?.[0]?.message?.content ?? 'No text content returned');
console.log({ finishReason: result.choices?.[0]?.finish_reason, usage: result.usage });

Run node first-request.mjs on your server. Add an output limit only after checking which parameter the selected model accepts; token-limit and sampling parameters are not universal across models. The timeout is a client policy, not a server latency guarantee.

Verify more than HTTP success

Inspect finish_reason before treating text as a complete answer. A length-limited response may need a smaller input or a reviewed output allowance. An empty text field may contain another response type or indicate a refusal; retain the native response for diagnosis without logging secrets or sensitive prompts.

LLM responses use the selected native protocol. They do not return the media pending task envelope and should not be polled through /v1/tasks. If the connection times out, do not assume generation was cancelled or free: investigate usage before resubmitting.

Continue with protocol selection, errors and retries, or the existing GPT-6.1 Sol guide. The model guide is an integration reference, not a guarantee of current availability.