Skip to content

VectorDash 0.3.42

Choose a tag to compare

@github-actions github-actions released this 19 Oct 21:23
1c01950

New features

  • Feature: Reverse image search (BETA) finds visually similar indexed files and retained clipboard images from a pasted or dropped image using compatible EmbeddingGemma 2 visual vectors.
  • Feature: Optional automatic image search submits pasted or dropped images in Search and Clipboard, with manual Enter-to-search remaining the default.
  • Feature: AI Local Search (BETA) lets tool-capable chat models search indexed file names and content snippets through an optional toggle beside Web search.
  • Feature: AI chat settings offer default Web access and separate screen-area capture Web/response-length preferences for the attached request.
  • Feature: Appearance settings provide independent start/end working-trail colors for each search mode and microphone recording, with reset defaults.
  • Feature: New controls are translated across all 13 supported interface languages.

Fixes since 0.3.41

  • Fix: Mapped network-drive PDF and image previews resolve the share path when elevated-process drive mappings are unavailable.
  • Fix: Mixed-DPI window dragging follows the Windows-suggested position instead of pushing the window back to the previous monitor.
  • Fix: Summoning a hidden search popup restores its saved position after wake or display changes.
  • Fix: Clipboard settings and their controls use the available window width.
  • Fix: Web result titles, links and descriptions stay beside thumbnails without overlapping them.
  • Fix: Image queries search all retained compatible clipboard image vectors instead of only the latest 500 images.
  • Fix: Enter submits image queries in Search and Clipboard, including a queued request while the image is still preparing.
  • Fix: Image result grids keep the first selected row visible while the popup expands.

Testing and beta features

  • Testing: 295 native tests and nine image/AI GUI scenarios passed locally, including manual Enter, queued Enter, automatic search in both modes, filters and first-row visibility.
  • Testing: The preceding beta build also passed nine AI capture/layout scenarios and six trail/settings/monitor scenarios, and a live 768-dimensional EmbeddingGemma 2 image-similarity check ranked a cropped test image against its original correctly.
  • BETA: Reverse image search, AI Local Search, Windows 10 support, camera RAW, interface localization, FP16 semantic caching and Rank Calibration need additional real-world testing.

Image queries require compatible visual embeddings; Quick-only and OCR-only entries have no visual vectors. Existing compatible visual vectors do not require rebuilding. Remote image queries send the selected image to the configured embedding server, and AI Local Search sends returned indexed snippets to the configured AI server.

Setup EXE and portable ZIP are available below, with SHA256SUMS.txt for download verification.