This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

dsh-vision-router

— source: plugin_wiki.wiki_content

Install & verify

dsh plugin --profile web add dsh-vision-router

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

• Why this exists
• How it compares
• Design lineage
• Acknowledgements
• Quick start

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

Do I need additional configuration after installation?

No, you don't. The plugin comes with a bundled patch and has the built-in OVH anonymous free visual chain enabled by default. After installation, you can select the model group with "+ Auto Image Recognition" in the bottom right corner of the chat page to send images.

Can I use it without an API Key?

Yes. The default chain includes 5 OVHcloud anonymous visual models (2 requests per minute per IP per model, totaling approximately 10 requests per minute), requiring no registration and no Key; if you feel the quota is insufficient, you can at any time connect to free tiers such as Zhipu, Alibaba Bailian, Intern AI, Groq, Google AI Studio, NVIDIA NIM, OpenCode Zen, OpenRouter, etc.

Why does the chat page say "Current model does not support images"?

The plugin deliberately does not modify the original model group; you need to switch to the model group with "+ Auto Image Recognition" in the model selector in the bottom right corner of the chat page to send images; sending images while the original plain text group is selected will be blocked by the host.

Which platforms are supported?

Cross-platform; vision_screenshot uses system screenshot capabilities on macOS/Windows, Linux requires ImageMagick's import or scrot; vision_html_screenshot requires Chrome/Chromium/Edge.

What visual tools are available?

There are 13-14 default mounts: vision_describe (image Q&A), vision_ground (pixel positioning), vision_detect (element listing), vision_crop (cropping), vision_pixel_diff (pixel comparison), vision_colors (color picking), vision_ocr (text transcription), vision_trace (SVG vectorization), vision_extract_foreground (background removal), vision_present (image presentation), vision_materialize (attachment saving), vision_html_screenshot (HTML screenshot), vision_long_screenshot_ocr (long screenshot transcription), plus vision_bootstrap for structured pre-recognition; the privacy-sensitive vision_screenshot is disabled by default, becoming the 14th when enabled.

Does it support purely local offline recognition?

Yes. After enabling localOllama or localLmStudio, the local visual backend ranks first in the HTTP visual chain, automatically degrading to the cloud chain if unavailable; after enabling instantDescribe, the first step of image processing is local recognition.

After upgrading, DSH reports "duplicate loader entry id: vision-router" - what should I do?

There are residual manual insert blocks from the v0.x era in the profile directory's cordis.patch.yml, which duplicate the bundled patch that comes with the plugin; delete the entire insert section, or rewrite it to overwrite lines by id.

How to uninstall?

Execute dsh plugin --profile web remove github:ysr666/dsh-vision-router; the wrapper routing, tools, skill cards, and settings cards will be removed together, while generated product files will be preserved.

— source: plugin_wiki.faq_json

Compatibility

• DSH: >=0.1.0-rc.6
• Node: >=22

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule