---
title: 'DeepSeek Harness Just Hit 35,000 AI Plugins?! 🤯'
source: 'https://youtube.com/watch?v=9h07pk0CjoU'
video_id: '9h07pk0CjoU'
date: 2026-11-19
duration_sec: 593
channel: 'Julian Goldie SEO'
---

# DeepSeek Harness Just Hit 35,000 AI Plugins?! 🤯

> Source: [DeepSeek Harness Just Hit 35,000 AI Plugins?! 🤯](https://youtube.com/watch?v=9h07pk0CjoU)

## Summary

DeepSeek Harness, an open-source AI agent framework, has rapidly grown to 35,000 third-party plugins and 1.54 million daily active users. The video explains the core philosophy behind this growth—'everything is a plugin'—and provides practical guidance on using the tool, including safety tips and future plans for an official plugin marketplace.

### Key Points

- **Plugin Explosion** [00:03] — DeepSeek Harness has grown to roughly 35,000 third-party plugins since launch, with about 60% of active users having built or installed at least one.
- **User Growth** [00:29] — Harness had 1.54 million daily active users as of November 4th, according to Technode.
- **What is Harness?** [00:43] — DeepSeek sums it up as 'Model + Harness = Agent'. The model does the thinking, while Harness gives it hands to use tools, read files, run commands, and work step by step.
- **Open Source and Stars** [01:11] — Harness is open source under the MIT license, with around 273,000 GitHub stars. It launched on August 13, 2026, alongside DeepSeek-V3 Pro.
- **The Big Idea** [01:37] — The real reason for 35,000 plugins is the rule: 'Everything is a plugin.' The agent loop, tool registry, session log, and even the connection layer are all plugins, allowing anyone to change anything.
- **Cordis Framework** [02:21] — Harness runs on a framework called Cordis, which allows plugins to be unloaded cleanly and to declare dependencies, enabling them to switch on/off automatically.
- **Official Marketplace Coming** [02:50] — DeepSeek announced an upcoming official plugin marketplace for easier publishing and discovery, though no launch date has been announced.
- **Current Capabilities** [04:05] — Harness can handle everyday work (documents, spreadsheets, slides), coding (exploring code, running tests), research (finding info, citing sources), and background tasks like batch processing and scheduled tasks.
- **Creator Mode** [04:49] — In Creator mode, you describe a plugin in plain chat, and the agent writes it for you, making plugin creation accessible to non-developers.
- **Modes and Directory** [05:26] — Harness has different modes (Standard, Minimal, PGC, Creator) which are just different plugin bundles. An independent directory called DeepSeek Plugins lists 30 repository-checked plugins across eight categories.
- **Who Can Use It?** [06:35] — Harness is in public preview worldwide, with a desktop app for macOS (Apple Silicon) and 64-bit Windows, or via npx command for developers. It works with other model providers and OpenAI-compatible endpoints.
- **Safety Tips** [07:31] — Tips include reading the safety notice first, not installing everything, checking plugin repositories/licenses/permissions, using profiles instead of forking, and trying Creator mode.

### Conclusion

DeepSeek Harness's explosive plugin growth is driven by its 'everything is a plugin' architecture, which empowers the community to customize and extend the tool. Early adopters who learn it now will have a competitive advantage as the ecosystem matures.

## Transcript

plugins for one AI tool in less than 3 months, and most people still haven't builders seeing that you're not? Stick with me because by the end of this actually is, why the plugin count just exploded, what DeepSeek says is coming
next, and the one habit you need before you install a single plugin. So here's DeepSeek-Harness account posted an update called Growing Together with our Plugin Community. In it, DeepSeek said third-party plugins have grown to
roughly 35,000 since launch, and about 60% of active users have built or installed at least one. That's not all. Technode reported that Harness had 1.54 million daily active users as of November 4th, and Pandaily
DeepSeek-Harness? You probably know DeepSeek makes AI models. Harness is not around a model. DeepSeek sums it up like this. Model + Harness = Agent. The model does the thinking. The Harness gives it hands. It lets the AI use tools, read
files, run commands, and keep working on a task step by step. It launched on August 13th, 2026 on the same day as DeepSeek-V3 Pro. It's open source under the MIT license. That means anyone can read the code, use it, and build on top
of it. And right now, the GitHub repo shows around 273,000 stars. To put that in perspective, launch coverage reported roughly 50,000 stars in the first 12 hours alone. And this didn't come out of nowhere. Reports say DeepSeek started
putting together a dedicated Harness team back in May with tools like Codex and Claude Code as the benchmark. So this is DeepSeek moving beyond just making models. They want to own the layer that turns a model into something
still doesn't explain 35,000 plugins. The real reason is one simple idea. and I'll show you exactly what that idea is in a second. Okay, so here's the big idea. DeepSeek built Harness on one rule. Everything is a plugin. And they
mean everything. Most AI tools have a fixed core. You can add a few extras on Harness doesn't work like that. The agent loop is a plugin. The tool registry is a plugin. The session log is a plugin. Even the part that connects to
don't like how one piece works, you don't have to rebuild the whole thing. even said in their update that built-in features can be replaced by third-party plugins. That's a big deal. It means the community can build something better
instead. All of this runs on a framework called Cordis. Mind Studio explains two things that make it special. First, when you unload a plugin, it undoes behind. Second, plugins can say what they depend on. Then they switch on and
off by themselves when those pieces show up or disappear even while Harness is plugin count blew up. When everything is a plugin, anyone can change anything and next? This is the part most people missed in the announcement. DeepSeek
plugin marketplace, a place where developers can easily publish their plugins and where you can easily find the ones you need. Quick pause before I show you the next part. We've got a 30-day DeepSeek Harness roadmap inside
the AI Profit boardroom. Instead of jumping between random tutorials, you actually learn how Harness and its plugins fit together. And there are four setup won't start, a plugin won't load, or you're just not sure what to try
through it live. It's built for exactly what you're learning in this video. You you stop guessing. Link in the description or go to ai
plugin security. And they're preparing a support program for plugin authors with technical discussions and promotion of their work. But here's the honest part. DeepSeek also said community plugins are still evolving. They vary a lot in how
mature they are. And as reports noted, there's no announced launch date for the shouldn't wait on it, okay? So what can Harness actually do today? According to few main areas. Everyday work like working with documents, analyzing
spreadsheets, drafting docs and creating slides. Coding like exploring code, running tests. Research like finding information, checking facts and citing sources. It can also run background tasks like batch processing files while
it tracks progress and it can run scheduled tasks. DeepSeek's own example is a weekly report that runs every Friday at 5:00 p.m. Beijing time. Then DeepSeek's page shows ones like agent teams, sub agents, scheduled tasks,
voice input, terminal, auto approval review and the agent loop itself. A few keep that in mind. And here's my favorite part, creator mode. In creator mode you describe the plugin you want in plain chat. Then the agent writes it,
you don't need to be a hardcore developer to make your own plugin, you just explain what you want. There's also a developer side. Harness shows execution traces and runtime details, so when a tool call fails you can see what
happened instead of guessing. And here's a wild example of how far the plugin idea goes. Mind Studio wrote about a DeepSeek research prototype called Super Agent. It watches GitHub and Feishu, picks up requests, works on
them in its own session and sends back a pull request and it's built entirely out it's a research prototype, not an official product, but it shows what's possible when every part is swappable. Harness actually has a few different
modes, Mind Studio, List, Standard, Minimal, PGC and Creator and each one is really just a different bundle of plugins. Again, everything is a plugin. There's an independent directory called DeepSeek Plugins. It's not run by
DeepSeek and it says so clearly. It lists 30 repository checked plugins across eight categories like tools, UI, sessions, skills and sandboxes. A few that adds image questions and text reading from images, a terminal
interface called Tianchu TUI, a custom tool plugin for sandboxed JavaScript tools and turn rewind which lets you roll back a conversation and your workspace. On GitHub plugin authors tag their projects with the topic DSH
browse what people are building, and some community plugins even bring discovery inside Harness itself. For example, there's one called DSH market can find, install, update, and remove plugins without leaving the tool. It's a
community project, not the official marketplace Deep Seek just announced, but it shows how fast people are filling the gap. So who can actually use this? Deep Seek says Harness is in public preview worldwide, and it's open to both
to click and go, there's a desktop app for Mac OS on Apple Silicon and for 64-bit Windows. If you're a developer, you install node.js, then run one command, npx at deepseek-ai/dshweb.
own computer and opens it in your browser. Or you can clone the source code from GitHub and build it yourself. Here's something a lot of people don't realize. Harness isn't locked to Deep Seek models. Mine Studio notes, it works
with other model providers and OpenAI compatible endpoints too. But one honest developer preview, and it's changing fast. The README warns there will be things you set up today might need fixing after an update. Also, Deep
pull requests. If you want to contribute, you do it through discussions or by building your own plugin. Okay, now the tips. These are the things that will save you the most headaches. Tip one, read the safety
headaches. Tip one, read the safety notice first. The GitHub README asks you to read the safety.md file before you run the project. Do it. It takes a few minutes. Tip two, don't install everything. With 35,000 plugins out
there, it's tempting to grab a bunch, but plugins are code that runs inside your agent. Start with one or two that fix a real problem for you. Tip three, fix a real problem for you. Tip three, check before you install. The Deep Seek
plugins directory gives some really good advice here. Look at each plugin's repository, license, and permissions. Then test unfamiliar plugins in a separate throwaway profile before you use them anywhere important. And
remember, a verified label there only means the repo checks out. It is not a full security audit. Tip four, use profiles instead of changing the core. Mind Studio recommends using profiles and patches instead of forking the whole
command that shows you exactly what harness is loading. That's super handy when something isn't working. Tip five, try creator mode. If you can't find the plugin you need, describe it in chat and let the agent build it. That's the whole
point of the system. You shape the tool around how you work. So that's deep seek harness. One simple idea, everything is a plugin and the community ran with it all the way to 35,000. Right now, there's a window where most
business owners haven't figured this out yet and that window won't stay open forever. The people who learn these tools early just get comfortable faster. daily tutorials so you're always learning the newest AI tools like
harness as they change. You get four coaching calls every week where you can a prompt library so you're not starting from a blank page every time and you get connect with other people learning the same tools. Link in the description or
go to aiprofitboardroom.com. If you want the full process, the SOPs and over 100 AI use cases like this one, join the AI Success Lab. It's our free community. Links in the comments and description. You'll get all the video
notes from this episode plus access to 94,800 members. The AI Success Lab is where you go to keep learning after the video everything we covered today so you can come back to them anytime. You'll find
step-by-step SOPs, a big library of AI use cases, and a community of people completely free to join so click the link, jump in, and I'll see you inside.
