Vision · DSH plugin

Key features

• ThestreamWithRegistration method for packaging llm/stream: request contains image blocks and the target model does not declare image input to be triggered (access door barriers before adaptor layer release by resolveModelInfo package, link)
• AutobridgeImages are automatically converted when they appear in a pure text model request without any need for the model or user to do anything
• Converting pictures to text description using visual models (descriptions with the ID tag "Photo results" and master models knowing clearly that content is derived from graphic models)
• Enter the description as a user/ message scenario-replace event append in the session log (compaction compression historical mechanism) - only visible to models, humans retrace original retention maps
• This request is rewritten with the same description, and the first round will be successful; embedded images returned by tools such as model call read image are also re-recognised and rewritten