Raw OpenAI-Compatible Fetch

If you already have an OpenAI-compatible HTTP client or a custom fetch wrapper, @vel4ai/ai-sdk is not required. Use the URL and headers returned by prepared execution to call /chat/completions. This is a Vela-managed relay call, so the relay records completion / failure and usage when available.

Non-Streaming

declare const prepared: import("@vel4ai/sdk").PreparedExecutionAllowed;
declare const messages: Array<{ role: "user"; content: string }>;

export async function callRelay() {
  const response = await fetch(prepared.openAiCompatibleChatCompletionsUrl, {
    method: "POST",
    headers: prepared.openAiCompatibleRequestHeaders,
    body: JSON.stringify({
      model: prepared.openAiCompatibleModel,
      messages,
    }),
  });
  const body = (await response.json()) as { usage?: unknown };

  return body;
}

Streaming

For streaming, return the relay response directly. The Vela-managed relay records the execution result asynchronously.

declare const prepared: import("@vel4ai/sdk").PreparedExecutionAllowed;

export async function proxyRelayStream(messages: unknown[]) {
  const response = await fetch(prepared.openAiCompatibleChatCompletionsUrl, {
    method: "POST",
    headers: prepared.openAiCompatibleRequestHeaders,
    body: JSON.stringify({
      model: prepared.openAiCompatibleModel,
      messages,
      stream: true,
    }),
  });

  return response;
}