Attachments & OCR
Key idea: Attachments are shown via proxied URLs; OCR results are session‑only (memory) and auto‑cleared.
Supported previews
- Images (PNG/JPG/GIF/SVG/WebP): zoom + pan; OCR on demand
- PDF: inline preview
- Text/HTML: inline preview
- Audio/Video: inline playback
- Unsupported: graceful fallback + download
OCR behavior
- Results cached in memory for the session only (typical cap: ~30 minutes, ~50 entries).
- No localStorage/disk writes for OCR text by default.
- “Visual Evidence” can be appended to summaries when useful.