AI Agent Hub
Back to plugins
⚙️

dsh-ds-vision-auto-route

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install litianshuo110/dsh-ds-vision-auto-route

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

Run dsh plugin install litianshuo110/dsh-ds-vision-auto-route in your DeepSeek Harness terminal to install the vision auto-route plugin from https://github.com/litianshuo110/dsh-ds-vision-auto-route.

About this plugin

In DeepSeek Harness, when your main model is text-only, pasting a screenshot or sending an image-bearing request either gets rejected outright or forces you to manually switch to a vision model. This breaks the flow, especially in multi-turn sessions that mix text and images.

The dsh-ds-vision-auto-route plugin solves this by automatically detecting image-bearing turns inside the agent request waterfall and routing those calls to a configurable vision-capable model, while keeping the UI selection unchanged and leaving text-only turns untouched. Routing policies include turn-image (default, routes only when the current turn introduced an image), any-image (routes whenever the log has ever contained an image), and off, all driven by the durable log for full auditability and reconstructability.

Ideal for DeepSeek Harness users who rely on a text-only main model but occasionally need image understanding. It works out of the box on a source build and will be natively supported in future official releases. If you want to pick a model once and let images be handled transparently by a vision fallback, this plugin is exactly that.

Use Cases

  • Paste a screenshot while using a text-only main model and it auto-routes to a vision model
  • Mix text and image inputs in multi-turn conversations without changing the model selector
  • Image requests go to a dedicated vision model while text requests stay on the selected model

Best For

  • Developers using a text-only main model who occasionally need image understanding
  • Harness users who want a stable model selection with background multimodal adaptation
  • DeepSeek Harness source-build developers needing vision routing support