Vision & Multimodal

dsh-vision-lmstudio

tiankundai/dsh-vision-lmstudio · ★ 1

tiankundai

Local LM Studio vision recognition plugin for the DSH (DeepSeek Harness) Web GUI — recognize, describe, and OCR local images or clipboard screenshots with a local vision model.…

License
BSD-3-Clause
Added
2026-08-23
Maintenance
active
Install kind
github

No manifest check is available; listing does not imply compatibility or security review.

安装

dsh plugin --profile web add github:tiankundai/dsh-vision-lmstudio

GitHub 源码安装:首次需按提示配置 allowBuilds 构建授权后重试

For install and environment setup, plus plugin development tutorials, see DSH community docs ↗

README

Source README · body: en · Read Chinese README

Table of contents
  1. Features
  2. Requirements
  3. Install
  4. Configuration
  5. Repository layout
  6. License

English | 中文

Local LM Studio vision recognition plugin for the DSH (DeepSeek Harness) Web GUI — recognize, describe, and OCR local images or clipboard screenshots with a local vision model. All inference happens on your own machine; images never leave it.

Features

  • Agent tools (registered host-side, available in every session)
    • lmstudio_vision — recognize a local image file by absolute path
    • lmstudio_clipboard — recognize the current clipboard image (screenshot / copied image), no file path needed
    • lmstudio_models — list loaded models and mark the most recently loaded one (the auto-selected one)
  • Settings panel: "Settings → LM Studio Vision" — configure server URL / model / default prompt / temperature / max tokens, refresh the model list, and one-click recognize a local image or the clipboard
  • Automatic model selection — without an explicit model, the most recently loaded model is used (the first entry of the server's model list; the native /api/v0/models endpoint is filtered by state, so models that are JIT-visible but not loaded never get picked). No specific model is preset.
  • Persistent config — ~/.dsh/dsh-lmstudio-vision.json (atomic writes), shared by the panel and the agent tools

Requirements

  • DSH (DeepSeek Harness) with the web profile
  • LM Studio with a vision model loaded and the local server started (Developer → Start Server, default port 1234)
  • Windows for the clipboard tool (uses Windows PowerShell 5.1 Get-Clipboard -Format Image)

Install

pnpm install
pnpm build
dsh plugin --profile web add link:<path-to-this-repo>/packages/dsh-lmstudio-vision

Then restart DSH (dsh web) and refresh the page. dsh plugin appends the package to dsh.profile.bundles automatically (the package declares dsh.bundle).

Configuration

Configurable from the settings panel or by editing ~/.dsh/dsh-lmstudio-vision.json:

{
  "baseUrl": "http://127.0.0.1:1234",
  "model": "",
  "prompt": "Describe this image in detail.",
  "temperature": 0.2,
  "maxTokens": 1024
}

Leave model empty to auto-select the most recently loaded model, or set an explicit model id to override it.

Repository layout

  • packages/dsh-lmstudio-vision/src/index.ts — host half: LM Studio engine, API routes, agent tools, system-prompt announcement
  • packages/dsh-lmstudio-vision/src/client/ — browser half: the settings panel
  • packages/dsh-lmstudio-vision/cordis.patch.yml — bundle patch plugin row
  • shared/tsdown.client.ts — build preset (follows the dsh-web-ui clientBundle convention)

License

BSD-3-Clause