Feature
Image attachments.
Curtis.ai supports vision — drag a screenshot in, paste a photo, or pick from a file. The model reads it and responds.
Curtis.ai supports vision — drag a screenshot in, paste a photo, or pick from a file. The model reads it and responds.
Three ways to attach
| Method | How |
|---|---|
| Drag-and-drop | Drag any image file onto the chat panel. A dashed accent outline confirms the drop target. |
| Paste | Ctrl+V while focused in the input box. The image on your clipboard attaches directly. |
| Paperclip | Click the paperclip icon to the left of the input box. Opens the OS file picker, multi-select supported. |
Attached images appear as 64×64 thumbnails in a strip above the input. Click the × on any thumbnail to remove it before sending.
Storage
Images are saved as real files in your vault's attachments/ folder (created if missing). Filenames are timestamped (Pasted image 2026-07-20-15-01-23.png) and unique.
The conversation file stores the vault path of each attachment, not the base64 bytes. That keeps conversation notes small and human-readable, and makes the images visible, auditable, and reusable in other notes.
At send time, Curtis.ai reads the bytes back from the vault and encodes them as a base64 data URL for the provider's API.
Vision-capable models
Look for the 👁 vision pill in the model picker — that model can read images. Models without it will reject image content.
If you send an image to a non-vision model:
- Curtis.ai detects the rejection automatically
- Shows a Notice: "This endpoint rejected the image — retrying as text only"
- Retries the message with images stripped, so you still get a response
For fully local vision, use minicpm-v via Ollama.
Size limits
- 8 MB per image — larger files are skipped with a Notice
- Auto-resize cap — the input textarea grows up to 320px before scrolling internally
For huge images, downscale before attaching. Most providers reject images above ~20 MB anyway.
Images in transcripts
When you /save-all or save a single message as a note, attached images are embedded as Obsidian wikilinks:
## 🧑 You
What's in this screenshot?
![[Pasted image 2026-07-20-15-01-23.png]]
The saved note renders them inline — same as if you'd pasted them directly into the note.
Click-to-zoom
In the chat, attached images render at max 180×180 in the user's message bubble. Click to open full-size in Obsidian's native image viewer.