What it does
Vision Router gives a text-only agent a set of pixel-level tools rather than a single describe-this-image call. It bundles a free vision chain that needs no key, then adds question answering, grounding, cropping, pixel diffing, colour sampling, OCR, SVG tracing, cutout and screenshot capture, so image work happens inside ordinary tool-calling turns.
Key features
- Free built-in vision chain with no API key required
- Grounding that returns coordinates for a described element
- Crop and zoom so fine detail stays readable
- Pixel diff for visual regression between two screenshots
- Screenshot capture feeding straight into the next tool call
Install
dsh plugin --profile web add dsh-vision-routerCompatibility notes
No Python dependency, which keeps installation simple on fresh machines.
