// Groq as a remote backend -- OpenAI-compatible chat completions on
// LPU hardware, notable mainly for raw speed (observed far faster
// token generation than either local Ollama or Gemini's free tier) and
// a generous free quota (30 req/min, 1000/day at time of writing).
import { createOpenAiCompatBackend } from "./openaiCompatBackend.js";
import type { ChatBackend } from "./chatBackend.js";

function apiKey(): string {
  const key = process.env.GROQ_API_KEY;
  if (!key) throw new Error("Groq backend needs an API key: set GROQ_API_KEY in your environment or app/.env. Get a free key at https://console.groq.com/keys -- never paste the key itself into chat.");
  return key;
}

export function createGroqBackend(model: string): ChatBackend {
  // Resolved eagerly, not passed as a lazy callback -- callers (zz
  // work/--assist) call this before resolveCliBackendMode/
  // enterFullscreen, specifically so a missing key fails fast as a
  // plain CLI error instead of surfacing deep inside a TUI session on
  // the first draft attempt (confirmed live: that's exactly what used
  // to happen).
  const key = apiKey();
  return createOpenAiCompatBackend({
    name: `groq:${model}`,
    url: "https://api.groq.com/openai/v1/chat/completions",
    model,
    apiKey: () => key,
  });
}