dsh-vision-api-localorweb

DSH assistant ships without any image-understanding capability out of the box, yet many developers already have a multimodal model running locally via LM Studio, Ollama, or vLLM. dsh-vision-api-localorweb bridges that gap: point it at any OpenAI-compatible base URL, name the model, and your assistant gains a pair of local eyes with zero data leaving your machine.

Under the hood the plugin registers a single tool, local_vision, which reads an image from a local path, sends it to your vision endpoint, and returns a text description into the conversation. Two usage paths are supported: supplying an image file path in chat (works even when the dialogue model is text-only), or dragging an image straight into the dialog when the dialogue model itself accepts vision input. Leaving the base URL blank disables the feature entirely, and a three-field settings page keeps configuration completely config-file-free.

Ideal for anyone who has already deployed multimodal models such as Gemma, Qwen-VL, or LLaVA locally and wants DSH to describe what it sees without a single byte touching the cloud. MIT licensed, install in one line.