dddzzz123-dz/dsh-read-image-plugin

dddzzz123-dz★ 0JavaScript最后同步: 2026-08-22

在 GitHub 打开

Image input fallback for DeepSeek Harness with native multimodal model detection and Volcengine Ark vision.

README 摘要

DSH Read Image Plugin 中文说明 An image-input fallback for DeepSeek Harness (DSH) that routes images according to the selected model's declared capabilities: - Models whose inputModalities include image use DSH's native attachment flow. - Text-only models get a fallback image button backed by the configured Volcengine Ark vision model and the read image via ark tool. This keeps native multimodal models on their direct path while giving text-only DeepSeek models a practical way to inspect images. Features - Image picker and clipboard image paste in the conversation input. - Host-side upload endpoint with a 20 MB limit. - Local file paths and HTTP(S) image URLs. - Reuses the DSH volcengine provider, credential, model, and API protocol. - Supports OpenAI-compatible Responses and Chat Completions APIs. - Automatically hides the fallback UI for models that natively accept images. Layout Install or link pkg as @local/dsh-read-image in the Web profile, enable @local/dsh-read-image/client , and load the host module from cordis.patch.yml : Restart the DSH Web UI after changing the profile. DSH profile paths and loading details may vary by release, so follow the plugin-loading convention used by…

在 GitHub 查看完整 README →
内容/媒体arkdeepseekdeepseek-harnessdsh-pluginimage-readingmultimodalvisionvolcengine

分类