DeepSeek-Harness-Vision-Tools¶

⭐ 12 · ✅ 活跃 · 插件 · 近期 ⬆️ +1


类型	插件	分类	视觉与多模态
星数	⭐ 12	状态	✅ 活跃
作者	tonyd2wild	更新时间	2026-08-13
子分类	👁️ 视觉工具	能力	multimodal, vision

一句话介绍¶

视觉代理：任意文本模型 + 任意视觉模型即可让 DSH 看图。

详细介绍¶

Give your DeepSeek Harness (dsh) eyes: keep your text model as the brain, and let a local vision model do the seeing. Any text model, any vision model, on Mac or Windows/PC. The model driving dsh is usually text-only (DeepSeek, dense Qwen, Llama, Mistral). Hand it an image and dsh refuses before it sends, naming the model. Declaring input: [text, image] to force it through is worse: the endpoint then answers 400 "not a multimodal model" mid-turn, after the message is already durable, and the session retries forever. So there are two doors an image can come through, and this repo ships a mechanism for each. They are complementary, not primary-and-alternative. Why both. The tool cannot serve a chat attachment: for

✨ 核心特性¶

parameters is a property map, not raw JSON Schema. `{ path: { type:
output must exist, with a schema and a render. defineTool reads

📦 安装¶

npm i -g @deepseek-ai/dsh@0.1.0-rc.6

🚀 快速开始¶

SessionCreateError: session create failed: agent-preset-invalid:
agent-presets: preset "<name>" failed to mount:
failed to apply loader entry tool-vision

📚 更多信息¶

Quick start

git clone https://github.com/tonyd2wild/DeepSeek-Harness-Vision-Tools cd DeepSeek-Harness-Vision-Tools cp .env.example .env # point the endpoints at YOUR hosts ./setup.sh # brings up a local vision server (+ the proxy with RUN_PROXY=1) Door 1, the proxy (chat attachments): python3 shim/vision_shim.py --port 8900 \ --upstream http://127.0.0.1:8000 \ --vision-url http://YOUR_FAST_VISION_HOST:808

Configuration

Proxy (env, or the matching CLI flag which overrides it): Tool (plugin config block or env): Full example: .env.example. ---