AI 分析

核心用途是让纯文本模型也能“看懂”图片。适合经常需要向纯文本模型发送截图的用户，插件会自动调用后台的多模态模型将图片转译为详细的文本描述，从而实现无缝的跨模态对话。

套件

dsh-autovision

版本

0.1.0

Usage

Paste an image into any session and send — the text model receives a faithful text transcription instead of the raw image.
Ask the model to read a file — the model may call autovision_read_image with a file path (and its own instruction) and act on the result.

Setting	Meaning
`defaultVisionModel`	Multimodal model used for transcription (from your LLM providers). No vision model → transcription degrades to a fixed placeholder instead of crashing.
`prompt`	Optional custom instruction for the vision model. Empty → an open-ended description prompt (text, colors, shapes, UI elements, layout, state).
`targetProviders`	Optional whitelist of providers to wrap (default: all).

Setting	Meaning
`defaultVisionModel`	Multimodal model used for transcription (from your LLM providers). No vision model → transcription degrades to a fixed placeholder instead of crashing.
`prompt`	Optional custom instruction for the vision model. Empty → an open-ended description prompt (text, colors, shapes, UI elements, layout, state).
`targetProviders`	Optional whitelist of providers to wrap (default: all).