Bespoke

Routes prompts to the right Gemini model and streams tokens back over SSE.

const fastMode = init || ["start", "services"].includes(step);
const miniModel = getFastModel();
const chatModel = fastMode ? miniModel : (env.GEMINI_MODEL || SMART_MODEL);

const stream = new ReadableStream<Uint8Array>({
  start: async (controller) => {
    const enc = new TextEncoder();
    const write = (obj: any) =>
      controller.enqueue(enc.encode(`data: ${JSON.stringify(obj)}\n\n`));

    const baseReq = {
      model: chatModel,
      messages: [{ role: "system", content: system }, ...history],
      stream: true,
      temperature: fastMode ? 0.2 : 0.4,
      max_tokens: fastMode ? 512 : 1000,
    };

    const completion = await openai.chat.completions.create(baseReq);
    for await (const part of completion as any) {
      const delta = part?.choices?.[0]?.delta;
      if (delta?.content) {
        assistantText += delta.content;
        write({ type: "text", content: delta.content });
      }
    }
  },
});