ErlichLiu

ErlichLiu / deepseek-vision

Public

为 DeepSeek v4 系列补齐视觉理解、联网搜索与 Anthropic / OpenAI 兼容接口的代理服务

43
8
85% credibility
Found May 30, 2026 at 43 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

deepseek-vision is an open-source proxy service that enhances DeepSeek's text-only AI models with image understanding and web capabilities. It works by intercepting requests that need vision or web features, processing them through additional AI services, and passing the results back to DeepSeek. The project includes a user-friendly web dashboard for configuration, supports both Anthropic-style and OpenAI-style API connections, and can be deployed via Docker or run directly. It includes security features like SSRF protection for web fetching and rate limiting for the admin interface. The project is MIT-licensed and maintained by Proma, a general-purpose AI agent project.

How It Works

1
💡 You discover a limitation

You want to use DeepSeek for your AI projects, but realize it can't see images or search the web like other AI assistants can.

2
🔌 You find the solution

deepseek-vision is a free tool that acts like a bridge, adding image understanding and web access to your DeepSeek setup.

3
🚀 You launch the service

With one simple command, you start the proxy server on your computer. It runs quietly in the background, ready to help.

4
⚙️ You set up your keys

You open a friendly web page where you enter your DeepSeek key and optionally keys for image recognition and web search. Everything has clear labels and helpful hints.

5
You connect your favorite AI tool
🤖
Claude Code or similar agent

Point your AI coding assistant to the proxy and start asking about images or current topics

💬
Chat interface or SDK

Use any OpenAI-compatible app with your DeepSeek setup and enjoy vision and search features

6
🖼️ Everything just works

Send an image and ask questions about it. Ask about news or recent events. The proxy handles all the complexity behind the scenes.

✨ Your AI project is complete

You've transformed DeepSeek into a full-featured assistant that can see, search, and fetch — all running privately on your own setup.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 43 to 43 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is deepseek-vision?

deepseek-vision is a Python proxy service that fills the gaps in DeepSeek's text-only API. It adds image understanding, web search, and web fetching capabilities so you can use DeepSeek with tools like Claude Code that expect those features. The proxy exposes both Anthropic Messages API and OpenAI Chat Completions endpoints, meaning any client built for those APIs works with DeepSeek through this layer. It ships with a web-based configuration dashboard and runs as a single Docker container.

Why is it gaining traction?

DeepSeek's official API cannot handle images or browse the web, which cripples agent workflows that depend on those capabilities. This proxy solves that by intercepting image blocks and converting them to text descriptions via Qwen or GPT-4o, while also handling web search and fetch tool calls. The two-round search architecture (plan queries, execute in parallel, generate answer) avoids the token bloat of iterative loops. You plug in one DeepSeek key and any OpenAI or Anthropic-compatible tool works immediately.

Who should use this?

Developers running Claude Code, Cline, or similar agents who want DeepSeek as their backend. Teams using LangChain or Cherry Studio with DeepSeek models. Anyone building workflows that need vision or web access but prefer DeepSeek's pricing and capabilities over alternatives.

Verdict

This is a clever, well-scoped solution to a real pain point. The credibility score of 0.8500000238418579% reflects a small but active project with solid architecture. However, with only 43 stars, it is early-stage software. The documentation is bilingual and the Docker setup is straightforward, but test coverage and community support are minimal. Worth trying for the specific use case of adding agent capabilities to DeepSeek, but do not bet production workflows on it without evaluating the codebase yourself.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.