dsh-vision-plugin

In a DeepSeek Harness conversation, the main model can only read plain text. A stack-trace screenshot, a whiteboard sketch, or a product UI mock-up all vanish the moment you paste them in. dsh-vision-plugin closes that gap: upload or paste an image right in the chat box, and on send the plugin transcribes the image into a text description that is injected into the message so the primary model actually sees what is in the picture.

Two strictly routed transcription paths are provided: the vision button or Ctrl+V paste triggers the DashScope visual model (requires an API key and a model that accepts image input), while the OCR button runs Windows 10+ built-in offline OCR with zero network calls. The two paths never silently fall back to each other; on failure the plugin tells you exactly whether the model call or the OCR step went wrong, keeps the image data in the preview bar for an instant retry, and makes sure the message is never forwarded to the primary model.

Built for Windows developers and knowledge workers who share screenshots, error logs, diagrams, or design mock-ups with AI on a daily basis. Images live in memory only and are cleared the moment transcription completes, keeping things both convenient and private.