kutcr/dsh-image-transcription-plugin

README excerpt

DSH Image Transcription Plugin 中文 · Upstream: DeepSeek Harness · Topic A source release for the DeepSeek Harness image-transcription feature. It is implemented as two Cordis plugins: - @deepseek-ai/dsh-image-transcription converts image-bearing messages into durable visual transcriptions when the primary model is text-only. - @deepseek-ai/dsh-tool-image-recall registers recall image , allowing the primary model to ask the configured vision model to re-read a cited historical image. Features - Stable transcriptionId references bind text transcriptions to their original attachments. - Compaction adds a compact image index; original attachments remain stored in the session log. - recall image sends only the cited image batch and the current question to the visual route. - The Web UI shows elapsed transcription status and retains clear failures. Configure the image-transcription model This plugin requires an image-transcription model configuration before it can process images. 1. Open Settings → Models . 2. In Image transcription model , select the visual provider and model to use when the primary model is text-only. 3. Expand Configuration and enter the provider API key, endpoint, pro…

Category