@vcxmug/dsh-vision

Native vision (image recognition) for DeepSeek Harness agents via any Responses API-compatible multimodal model endpoint. Fills the gap while DeepSeek has no official vision support.

AI Analysis

核心用途是为 DSH 引入图像理解与识别能力。适合在官方尚未提供原生视觉支持时，需要让 AI 代理分析和描述图片的用户。

Package

@vcxmug/dsh-vision

Version

0.1.0

License

MIT

Last updated

Aug 13, 2026

Install (local, no npm publishing)

npm pack ./packages/dsh-vision
npm install -g ./vcxmug-dsh-vision-0.1.0.tgz

Add a pure mount-point row to your agent preset ($DSH_HOME/.agent-presets//agent.cordis.yml, or edit the preset in the Web UI: Settings → Agent presets):

- id: vision
  name: '@vcxmug/dsh-vision'

Configure in the Web UI: Settings → Plugins → dsh-vision (baseUrl, model, apiKey, apiKeyFile, detail, timeouts). Changes apply on the next tool call — no session restart.

Start a NEW session with that preset. The model sees vision_describe and vision_list_models and uses them whenever an image must be understood.

Install

README

Install (local, no npm publishing)