Skip to content

1.1.0

Choose a tag to compare

View all tags
@saorgage saorgage released this 21 Jun 18:34
· 3 commits to main since this release
1.1.0
502e2e2

Vision: see and act on an image in one turn

When your chat model can't see an attached image (e.g. DeepSeek), the turn now routes through your configured vision provider (Gemini, etc.) running the full tool loop — so it can see the image and act on it in a single message.

Added

  • save_attached_image — saves the actual pasted/attached image bytes into your vault and returns a clickable ![[link]]. So "save this screenshot and tell me what it says" works in one message.
  • Image turns now have the complete tool set (create/edit notes, search, etc.) driven by the vision model.

Fixed

  • Images were unusable on follow-up turns — the image was dropped after the first send, so follow-ups fell back to a model that can't see images and would claim no image was passed. Attached images are now sticky: the chip persists across turns; tap × to remove.
  • A stray "undefined" was injected into the message text whenever an image was attached.

Verified end-to-end against Gemini 2.5 Flash (OpenAI-compatible): vision + tools + streaming + multi-step tool results all work together.

Assets 5
Loading