dsh-plugin-image-input

When you are running a text-only backend like DeepSeek or GPT-4 base in DSH, pasting a K-line chart or a screenshot into the chat either gets rejected outright or produces a hallucinated response. dsh-plugin-image-input closes that gap: paste or drag an image into the input box and hit send — the plugin calls the vision API you configure, converts the image into a structured text description (canvas elements, percentage coordinates), and dispatches that description together with your own text to the language model, so it can reason about charts and screenshots without ever seeing pixels.

For multimodal models (qwen-vl, gpt-4o, etc.) the plugin stays completely out of the way and lets the native image channel handle things. You can also tap the Image-to-Text button next to the input box to insert only the description into your draft, edit the wording, and then send — handy when you want finer control over how the model interprets the image. Multiple images are converted one by one and bundled into a single message; nothing is lost in the process.

Built for developers who run non-multimodal backends in DSH yet frequently need the model to understand charts, UI screenshots, or annotated diagrams. Configure the endpoint once (baseUrl, model name, API key) and the plugin works with any OpenAI-compatible vision API. No local sandbox required — runs on Windows, Linux, and macOS alike.