What it does
Modlens is the vision bridge for models that cannot see. Instead of returning a prose description that a text model has to trust, it returns structured JSON evidence: recognised text with positions, layout regions, and semantic labels for what each region appears to be. The model can then reason over evidence, cite a region, or ask for a tighter crop.
Key features
- OCR output with per-block coordinates rather than a flat string
- Layout segmentation into headers, tables, code, forms and media
- Semantic labels that let a text model reference regions by name
- Works with pasted images and workspace file paths
- Structured output that survives downstream tool calls
Install
dsh plugin --profile web add modlensCompatibility notes
Very long screenshots benefit from tiling; check the plugin’s tile settings if fine print comes back garbled.
