Benjam1n77/dsh-kiroshi

dsh-kiroshi gives DeepSeek Harness vision through an external vision model so pure-text DeepSeek models can "see" images with no model switch: pasting an image auto-produces a one-line summary and the kiroshi_read_image tool reads the original on demand, with content-addressed dedup storage and in-chat thumbnails.

dsh-kiroshi gives DeepSeek Harness vision through an external vision model, so pure-text DeepSeek models can 'see' images with no model switch and no change of habit. Pasting an image has it auto-recognised and summarised, and asking for details makes the model read the original through the kiroshi_read_image tool. It is token-efficient, producing a one-line summary by default and a full read only on demand, deduplicates storage by content address and renders the sent images as thumbnails in the chat under your message.

Vision & Multimodal ★ 2
View on GitHub ↗

Install

npx -y @deepseek-ai/dsh plugin --profile web add github:Benjam1n77/dsh-kiroshi

Copied verbatim from the README's install block, which documents the primary npm form (npx -y @deepseek-ai/dsh plugin --profile web add dsh-kiroshi) and this GitHub form 'before npm publish'. The npm package dsh-kiroshi currently returns 404 on registry.npmjs.org (check 2026-10-14), so the documented GitHub form is shipped.

Compatibility

A DeepSeek Harness Web plugin that works on the stock host with no upstream patch; configuration lives in ~/.kiroshi/config.json (baseUrl, apiKey, model - the README example uses SiliconFlow with a Qwen3-VL vision model). It stores images content-addressed so the same image is stored once, and shows the images you sent as thumbnails under your message in the client half.

Details

Alternatives

54xkeee/dsh-vision · 54xkeee/dsh-youreyes · Cheng-cheng9669/dsh-deepseek-vision

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins