DeepSeek: what are you trying to do?
4150 projects collected from GitHub, Hugging Face, GitLab, npm and Docker Hub — each one scored for whether it actually works, not just whether it exists.
Projects
4150
Active
2005
Devices
443
Verified
2026-11-04
Pick your starting point
② Run it on your own server
172 projects · Ollama, llama.cpp, vLLM, SGLang
③ Run it on a device
443 projects · Jetson, Raspberry Pi, RK3588, ESP32
④ Connect it to your app
264 projects · LangChain, Dify, FastGPT, bots
⑤ Use an existing client
596 projects · desktop, web, mobile
Models on Hugging Face
438 entries · official and community
⑥ Extend Harness
297 picked, not a full directory
Why this is not a link directory
Every entry carries a health status derived from its last commit date, and flags for the things that waste your afternoon: a README too thin to install from, a missing licence, an archived repository. The credibility score combines stars, recency, README depth and licence.
Devices are the gap. Run DeepSeek on a Jetson, a Raspberry Pi,
an RK3588 board or an ESP32 and the documentation thins out fast. That is where most of the
unanswered questions are, and where we put most of our collection effort.
Machine-readable
GET https://deepseekbase.com/data/ecosystem.json