Back to home

zoahdev

dsh-vision

Give DeepSeek Harness eyes: analyze images with an OpenAI-compatible vision model via a vision_analyze tool.

Stars
0
Language
TypeScript
Created
Aug 17, 2026
Updated
Aug 17, 2026

Introduction

dsh-vision

Give DeepSeek Harness eyes: a vision_analyze tool that sends an image (local path or URL) to an OpenAI-compatible vision model and returns a text answer.

License: MIT

English | 中文

Install

dsh plugin --profile web add dsh-vision

Set apiKey (or OPENAI_API_KEY), or point baseUrl at a local OpenAI-compatible endpoint such as Ollama (http://localhost:11434) with a vision model.

Tool

ParameterTypeDescription
imagestringImage path or https/data: URL
questionstringWhat to ask (default: "Describe this image in detail.")

Config

KeyTypeDefaultDescription
apiKeystring(env)OpenAI-compatible API key
baseUrlstringhttps://api.openai.comAPI base URL
modelstringgpt-4o-miniVision model id
timeoutMsnumber120000Request timeout

Honest limits

  • Thin bridge; it needs a vision model (API key or a local endpoint).
  • Community plugin, not an official DeepSeek or OpenAI product.

中文

给 DeepSeek Harness 装上眼睛:vision_analyze 工具把图片(本地路径或 URL) 发给 OpenAI 兼容的视觉模型,返回文字描述。

dsh plugin --profile web add dsh-vision

配置 apiKey(或 OPENAI_API_KEY),或把 baseUrl 指向本地 OpenAI 兼容端点 (如 Ollama http://localhost:11434 + llava)。