Open Questions for Implementation Team

  1. Minimax vision endpoint: What is the exact URL and payload format for Minimax’s vision API? Does it support base64-encoded images, or only URLs?
  2. BlueBubbles attachment retrieval: What fields does BlueBubbles return for attachments? Does it provide a download URL, a file path, or base64 data? How do we authenticate attachment downloads?
  3. Image resizing: Should we resize images before sending to Minimax? What are Minimax’s size limits? Should we use PIL/Pillow or ImageMagick on the Mac?
  4. Cache format: JSON file or SQLite? JSON is simpler but slower for large caches. SQLite is more robust but adds a dependency.
  5. Privacy consent UI: Should the consent be a modal dialog in the PyGame UI, or a simple config flag that defaults to false? A modal requires new UI code.
  6. HEIC support: iPhones send HEIC images. Can Minimax process HEIC? If not, do we convert to JPEG using heif-convert or similar?
  7. Rate limiting: Should we implement a global rate limiter (e.g., max 1 description per 2 seconds) to avoid hitting Minimax limits?
  8. Description length: The prompt requests “one concise sentence” but the model may return longer text. Should we truncate at a hard limit (e.g., 120 characters) or let the model decide?