Preface¶

In DSH Web, multimodal models can directly receive images, while pure text models cannot process images natively. kbpoyo/dsh-image-bridge is a DSH Web plugin that provides image bridging capabilities in pure text model conversations, giving pure text models the opportunity to view images as well.

What This Is¶

kbpoyo/dsh-image-bridge is maintained by kbpoyo and is licensed under the MIT License. It solves the problem in DSH Web where “pure text models cannot directly send images.”

The plugin is not bound to any specific skill or tool. Whether a model can ultimately view images depends on its own visual skills/tool catalog.

Core Features¶

• Provides image bridging capabilities for pure text models
• When an image is pasted under a pure text model, a “Bridge Send Image” button appears; click it to send
• Pressing Enter directly to send an image also automatically rewrites it to a bridge marker
• Under multimodal models, the button is automatically hidden and images are sent natively
• When switching models during a session, the button visibility updates immediately
• Temporary images generated by bridging are automatically cleaned up on a timer

Installation and Activation¶

The following installation command applies specifically to the web profile:

dsh plugin --profile web add @kbpoyo/dsh-image-bridge@0.1.0

After installation, restart dsh web for the plugin to take effect.

You can also use the following command to confirm that the plugin configuration layer has been mounted:

dsh --profile web --dump-config

Applicability to other profiles is not provided.

Typical Usage¶

1. In a session, select a pure text model (such as deepseek-v4-flash), paste or drag in an image, and click “Bridge Send Image” to send it.

2. Pressing Enter directly to send an image also triggers automatic bridging, equivalent to clicking the button.

3. When OCR is needed, explicitly state it in your prompt, for example, “Extract the text in the image.”

4. After switching to a multimodal model (such as mimo-v2.5), the button automatically disappears and images pass through natively.

Applicable Scenarios and Notes¶

This plugin is suitable for scenarios where you need pure text models to view images in DSH Web. Please note the following before use:

• The plugin loads and runs with the current dsh process. It is recommended to review the source code and MIT License before installation.
• The plugin is not bound to any specific skill or tool. Whether a model can view images depends on its own visual skills/tool catalog.
• Custom provider models that do not declare input: ['text','image'] will be reported as pure text, and the bridge button will be displayed.
• DSH native limitation: after a session log contains native images, switching back to a pure text model is prohibited; you need to use /compact or start a new session.
• Image-containing messages sent from a parent agent to a child agent do not go through session.prompt and still go through the native gate.
• Bridged images are written to the hidden session workspace directory <cwd>/.dsh-image-bridge/ with a TTL of 24 hours; when the total size exceeds 512MB, the oldest files are cleaned up first.

Related Links¶

• Catalog page: https://www.skillhub.cn/plugins/kbpoyo/dsh-image-bridge
• GitHub: https://github.com/kbpoyo/dsh-image-bridge