AI Agent Hub
Back to plugins
🤖

dsh-modlens

Model Inference Updated 2026.08.15

Run the following command in DeepSeek Harness:

dsh plugin install YZz-S/dsh-modlens

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install YZz-S/dsh-modlens to install the plugin; source code is available at https://github.com/YZz-S/dsh-modlens

About this plugin

Text-only LLMs cannot perceive images, and that gap is one of the most common blockers when building multimodal workflows. dsh-modlens bridges the gap by converting an image into structured JSON evidence—OCR text, layout analysis, and semantic description—so a text-only model can read charts, inspect UIs, and interpret documents without switching to a multimodal model.

Its core strengths include multi-engine support (Gemini, Anthropic, Claude Code, Doubao / Volcengine Ark, and more) with per-call engine selection via a provider argument. When several engines are configured, the model proactively asks the user which one to use, removing the friction of manual switching.

It is well suited for developers who are already running DeepSeek Harness and whose primary model only handles text: one plugin install is all it takes to give an existing agent the ability to see, without replacing the entire model stack.

Use Cases

  • Let a text-only model extract data and trends from charts
  • Read UI screenshots and locate elements to drive automation
  • Convert document photos into structured text for downstream analysis

Best For

  • Developers on DeepSeek Harness whose primary model is text-only
  • Teams that need image understanding in existing agents without changing the model stack
  • Builders who want centralized management of multiple vision engines with per-call switching