deepseek-harness-SupportVisionModel¶
| 类型 | 插件 | 分类 | 视觉与多模态 |
| 星数 | ⭐8 | 状态 | ✅ 活跃 |
| 能力 | multimodal, vision | 更新时间 | 2026-08-14 |
📖 项目简介¶
DeepSeek Harness SupportVisionModel is a community fork of DeepSeek Harness. It keeps the upstream plugin architecture and adds a unified image-routing path for image-capable and text-only main models. This repository is maintained independently at TryDing-T/deepseek-harness-SupportVisionModel. It is not an official DeepSeek AI distribution.
✨ 核心特性¶
- Model catalog entries can explicitly declare text-and-image input support.
- A dedicated Vision settings page selects the auxiliary provider and model, maximum attempts, per-attempt timeout, and output-token limit.
- Image-capable main models receive the original image attachments.
- Text-only or unknown-capability main models receive a validated description from the configured vision model.
- Missing or exhausted vision routes degrade to explicit attachment references instead of rejecting a successful user or tool action.
- Web uploads, top-level tool image results, ordinary
read_image, and nested Code Moderead_imageshare the same router.
📦 安装¶
git clone https://github.com/TryDing-T/deepseek-harness-SupportVisionModel.git
cd deepseek-harness-SupportVisionModel
pnpm install
pnpm run build
pnpm dsh web
🚀 快速开始¶
git fetch upstream
git diff main..upstream/master