VeryInt/dsh-image-vision
Seamless image understanding for DSH: lets pure-text main models read pasted or dropped images — and Feishu/Lark images, downloaded via lark-cli and described — through a configurable vision model. It is a pure "everything is a plugin" implementation: it wraps a documented llm service method and listens to official agent/tool waterfalls, touching no packages/ files, so it runs wherever DeepSeek Harness runs. A verification section in the README walks through confirming the route end to end.
Install
npx @deepseek-ai/dsh plugin --profile web add github:VeryInt/dsh-image-visionREADME install, from GitHub (recommended). A local-path form is documented for development: clone the repo and run npx @deepseek-ai/dsh plugin --profile web add /path/to/dsh-image-vision. Replace web with whichever profile you use (web, headless, ...). If pnpm blocks a git dependency's build/prepare script, allowlist the exact key it prints under allowBuilds in the profile's pnpm-workspace.yaml and re-run. Requirements: Node.js >= 22.19.
Compatibility
Node.js >= 22.19. Works with both deployment styles — a source checkout (pnpm dsh web) and the installed release (npm i -g @deepseek-ai/dsh) — because it goes through the same dsh plugin manager. A vision model must be configured (ModelScope, SiliconFlow, or any OpenAI-compatible endpoint) via the plugin's plugin vision route; maxImagesPerMessage defaults to 4, and the Feishu/Lark bridge is optional (feishuImages, maxFeishuImages default 5).
Details
- Repo: VeryInt/dsh-image-vision
- Category: uncategorized
- Stars: 0
- Version: GitHub main (README-documented; npm name dsh-image-vision is owned by a different publisher — wangyang10/image-vision, not this repo — so no npm claim is made)
- Last push: 2026-08-17
- First seen: 2026-08-17
Recent updates
README documents how this differs from oil-oil/dsh-vision: this plugin stays entirely out of tree (no host code changes, no patching the built-in plugin) and routes through the public llm service seam instead.
FAQ
- How do I install dsh-image-vision?
- Run: npx @deepseek-ai/dsh plugin --profile web add github:VeryInt/dsh-image-vision (or add a local checkout path), replacing web with your profile name. Node.js >= 22.19 is required.
- Which vision provider does it use?
- Any vision model you configure for its plugin vision route — the README documents ModelScope, SiliconFlow and OpenAI-compatible endpoints; set API keys per provider and optionally cap images with maxImagesPerMessage.
- Does it modify my DeepSeek Harness installation?
- No — it is purely out of tree: it wraps the documented llm service method and listens to official agent/tool waterfalls, so no packages/ files are changed.