Function callConfigured

src/tools/vision/vision-analyze.ts:342–367

Send an OpenAI-compatible vision request to the user's configured provider (Gemini's * OpenAI endpoint, OpenRouter, a custom-baseURL openai, …) whose primary/sub-agent model * can see. This is the "you already configured a vision-capable model — use it" backend.

(base64: string, mime: string, prompt: string)

 *  OpenAI endpoint, OpenRouter, a custom-baseURL openai, …) whose primary/sub-agent model
 *  can see. This is the "you already configured a vision-capable model — use it" backend. */
async function callConfigured(base64: string, mime: string, prompt: string): Promise<string> {
  const v = resolveConfiguredVisionModel();
  if (!v) throw new Error('No configured vision-capable model (primary/sub-agent is text-only or its provider has no key).');
  const res = await fetch(`${v.baseUrl}/chat/completions`, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json', 'Authorization': `Bearer ${v.apiKey}`, 'User-Agent': 'qodex-cli' },
    body: JSON.stringify({
      model: v.model,
      max_tokens: 1024,
      messages: [{ role: 'user', content: [
        { type: 'text', text: prompt },
        { type: 'image_url', image_url: { url: `data:${mime};base64,${base64}` } },
      ] }],
    }),
  });
  if (!res.ok) {
    const t = await res.text().catch(() => '');
    throw new Error(`Configured vision backend HTTP ${res.status}: ${t.slice(0, 300)}`);
  }
  const body = await res.json() as { choices?: Array<{ message?: { content?: string } }> };
  const content = body.choices?.[0]?.message?.content ?? '';
  if (looksLikeTextOnlyRefusal(content)) {
    throw new Error(`Configured model '${v.model}' responded as if it can't see images — its weights may be text-only despite the id.`);
  }
  return content;
}

function looksLikeTextOnlyRefusal(text: string): boolean {
  const lower = text.toLowerCase();