AI Agent Hub
Back to plugins
🤖

dsh-vision-fallback

Model Inference Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install 1HelloMan1/dsh-vision-fallback

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

To install the dsh-vision-fallback plugin in DeepSeek Harness, run the command dsh plugin install 1HelloMan1/dsh-vision-fallback, with the source code available at https://github.com/1HelloMan1/dsh-vision-fallback.

About this plugin

Powerful coding models like DeepSeek V4 Flash are strictly text-only, causing frequent errors when users try to drop or paste images directly into the chat interface. dsh-vision-fallback is designed to bridge this gap by intercepting images at the request layer. It automatically forwards them to a configured vision model (such as mimo-v2.5) for analysis and converts the resulting text observation into hidden context for your main model. This "silent" processing ensures that your original image remains visible in the UI while the underlying text-only model gains the ability to "see" the visual content without the need for manual file saving or complex tool invocation.

Use Cases

  • Process images directly in chat with text-only models without saving files.
  • Automatically handle persistent image attachments in MCP/ACP/PTC modes.
  • Maintain original UI while allowing the main model to see visual context via hidden layers.

Best For

  • Developers using DeepSeek V4 Flash or other text-only models.
  • Users needing to handle multimodal MCP/ACP attachments in DeepSeek Harness.
  • End users who want to maintain the original chat interface style while automatically processing images.