Function detectLmStudioContextWindows

src/setup/model-detector.ts:253–276

(
  ports: number[] = [1234, 8080],
)

 * isn't reachable (older LM Studio, llama.cpp, etc.).
 */
export async function detectLmStudioContextWindows(
  ports: number[] = [1234, 8080],
): Promise<Record<string, number>> {
  const out: Record<string, number> = {};
  for (const port of ports) {
    const r = await fetchWithTimeout(`http://127.0.0.1:${port}/api/v0/models`);
    if (!r || !r.ok) continue;
    try {
      const body = await r.json() as {
        data?: Array<{ id: string; max_context_length?: number; loaded_context_length?: number }>;
      };
      if (!Array.isArray(body.data)) continue;
      for (const m of body.data) {
        // Prefer the loaded window (what's actually usable right now); fall back
        // to the model's max. Either is vastly better than a hardcoded 32768.
        const ctx = m.loaded_context_length || m.max_context_length;
        if (m.id && typeof ctx === 'number' && ctx > 0) out[m.id] = ctx;
      }
    } catch {
      /* ignore parse errors — heuristic will cover it */
    }
  }
  return out;
}

/**
 * Recommend a primary model from the detected list.