A Chinese AI Agent App Just Hit 1.5 Million Daily Users — and Nobody in the West Noticed

By Shakeel Ahmed — AI Growth Lab, November 9, 2026

DeepSeek's open-source agent app, called Harness, reached 1.54 million daily active users by November 4, 2026 — less than three months after its first version launched on August 13. At the end of September it had about 271,000 daily users. That is nearly six times more in about five weeks, and it makes Harness one of the fastest-growing AI products of the year. The numbers come from DeepSeek itself, disclosed on X on November 8, and they have not been independently verified. But even with that caveat, the story matters: while Silicon Valley argues about AI agents, millions of people are already letting one run their work.

This article explains what Harness does, why it grew so fast, what it says about the race between open-source and closed AI agents, and what you should know before letting an agent act on your behalf.

1. The numbers

DeepSeek disclosed the figures on Sunday, November 8, on X. Harness had 1.54 million daily active users as of November 4, and it was consuming about 92 trillion tokens a day — roughly 60 million tokens per active user per day. That token number is the quiet bombshell in the announcement. Tokens are the units of information a model processes, and agents burn through them as they plan, call tools, and execute tasks. It is among the first hard evidence of how steeply AI agents escalate demand for computing power.

The growth curve is steep. The first version, v0.1, launched on August 13, 2026. The desktop app followed on September 29. Daily users stood at about 271,000 on September 30, shortly after the desktop release. By November 4 they had reached 1.54 million. The count includes only users connecting to DeepSeek's official API, so people who connect third-party APIs are not counted — the real total could be higher.

The plugin ecosystem is growing with it. Harness has about 35,000 third-party plugins, and roughly 60 percent of active users have created or installed at least one. DeepSeek says it plans an official marketplace where developers can publish plugins and users can discover them.

One detail deserves care. The 1.54 million figure counts Harness users on the official API. Harness is a separate product from the well-known DeepSeek chat app, which has a far larger user base. Do not confuse the two numbers.

2. What Harness actually does

Harness is an agent runtime — a piece of software that organizes models, tools, skills, sessions, sandboxes, and storage so an AI agent can carry out multi-step work with limited human supervision. In plain words, it is not a chatbot you talk to. It is a worker you give a job to.

It handles tasks such as coding, file analysis, and research. You describe the outcome you want, and Harness plans the steps, calls the tools it needs, and works through them. Its plugins are replaceable and composable, meaning users can assemble different workflows from small pieces — one plugin for searching files, another for running code, another for drafting a document — and the agent chains them together.

That design is why the plugin count matters. Harness is open-source, so developers can build and share plugins without asking DeepSeek's permission. The 35,000 plugins already built are the ecosystem, and the planned official marketplace is DeepSeek's attempt to organize it the way an app store organizes apps.

3. Why it grew so fast

Three things drove the surge. First, the desktop release on September 29 put Harness on people's actual work computers, not just in a browser tab. The jump from 271,000 users at the end of September to 1.54 million by November 4 starts right there.

Second, it is open-source. In China, the AI office market is already crowded — Tencent's WorkBuddy had 32.35 million monthly active users as of September, Baidu's Dazi had 9.77 million, Alibaba's Qianwen Office had 7.9 million, and ByteDance's Doubao Work had 2.05 million, according to market data from QuestMobile. Harness is not competing with those as a chat app. It is competing as the open infrastructure underneath agent work, and open infrastructure spreads fast when developers can extend it for free.

Third, the timing is right. AI agents are the industry's next frontier after chatbots. OpenAI and Anthropic are both pushing agents that act on your behalf, and DeepSeek — the Hangzhou startup that jolted Silicon Valley last year with its low-cost reasoning model R1 — is staking its claim on the agent layer rather than the chat layer.

There is one honest catch. Users usually need to pay for the API costs when using models on Harness, and the threshold for free use is much higher than on several competing products. Growth this fast, with users paying their own compute bills, is a stronger signal than growth bought with free credits.

4. Open-source vs closed: what the agent race looks like now

Harness is the clearest test yet of whether open-source agents can beat closed ones. The closed model — OpenAI's and Anthropic's agents — is polished, controlled, and distributed through one company's app. The open model — Harness — is rougher, cheaper to extend, and distributed through thousands of developers building plugins.

History favors open infrastructure in the long run. The web, Linux, and most of the software the internet runs on are open. But agents are different from web servers in one way that matters: they act. They send emails, change files, spend money, and make decisions with real consequences. That makes trust, safety, and accountability part of the product — and those are the areas where closed systems, with one company answerable for what happens, have an easier story to tell.

DeepSeek's answer is the plugin marketplace and better plugin API stability, compatibility, and security — all items the company says it is working on. The race is not decided. But 1.54 million daily users in under three months means the open side of it can no longer be dismissed as a hobbyist project.

For the other side of this race, read our coverage of the warning Apple just gave about agents that act without asking: Apple Just Warned: AI Agents Could Spend Your Money Without Asking: https://sheen6869.wixsite.com/shakeel-ahmed-ai-blo/post/apple-just-warned-ai-agents-could-spend-your-money-without-asking

5. What Western users should know before trying an agent that acts for them

Most Western readers cannot simply download Harness and start using it today. It is built for the Chinese market first, and its documentation, plugins, and support are centered there. But the questions it raises apply to every agent, from any company, and they are worth asking before you hand any agent your passwords, files, or calendar.

First, know the difference between reading and acting. An assistant that reads your email and summarizes it cannot hurt you much. An agent that sends email, moves files, or books things on your behalf can. Harness acts. So do the agents OpenAI and Anthropic are shipping. Before you use one, check what it is allowed to do without asking you — and whether you can set an approval step for anything that costs money or cannot be undone.

Second, know who pays for the compute. Agents burn tokens at a rate chatbots never did — Harness averages about 60 million tokens per user per day. Somebody pays for that. With Harness, it is usually you, through API costs. With closed agents, the cost is buried in a subscription. Either way, an agent that runs all day is not free, even when the app is.

Third, know where your data goes. Harness counts only users on DeepSeek's official API, which means your tasks run through DeepSeek's servers. The same is true of every cloud agent. If your work involves client data, health information, or anything under a confidentiality agreement, check the terms before you paste it in.

Fourth, expect mistakes. Agents are more capable than chatbots and more wrong in more expensive ways. A chatbot gives you a wrong answer you can ignore. An agent can send the wrong email, delete the wrong file, or book the wrong flight. The safe way to start is with low-stakes tasks — research, drafting, organizing — and a human check on everything else.

Voice agents are one place this is already visible in the West: when conversations get long and complicated, voice agents beat chat, and the people who can deploy them are in demand. Read how that works here: Voice AI Just Beat Chat at the Hard Conversations — Here's What That Means for Your Business: https://dev.to/shakeelahmed/voice-ai-just-beat-chat-at-the-hard-conversations-heres-what-that-means-for-your-business-1gn2

6. The bottom line

DeepSeek Harness went from 271,000 daily users to 1.54 million in about five weeks, on an open-source agent that lets people build their own plugins and put AI to work. The figures are DeepSeek's own, not independently verified, and the product is aimed at China first. But the direction is unmistakable: the agent era is not coming. It is already here, it is growing faster than the chatbot era did, and it is growing fastest where it is open.

The question for everyone else — users, developers, and the Western companies racing to catch up — is not whether agents will act on our behalf. It is who we trust to build the ones that do, and what we demand before we hand them the keys.

If you want the wider picture of governments trying to put rules around AI before it outruns them, read our coverage of the US bill that would ban the most powerful AI outright: A New US Bill Could Ban Powerful AI — Here's What It Actually Says: https://sheen6869.wixsite.com/shakeel-ahmed-ai-blo/post/a-new-us-bill-could-ban-powerful-ai--heres-what-it-actually-says

Sources: Emergent — DeepSeek Harness Hits 1.54M Daily Users in 3 Months: https://emergent.sh/news/deepseek-harness-hits-1-54m-daily-users-3 — TradingView / Seeking Alpha — DeepSeek Says Agent App Tops 1.5M Daily Users, Plans Plugin Marketplace: https://www.tradingview.com/news/seekingalpha:df5e95257094b:0-deepseek-says-agent-app-tops-1-5m-daily-users-plans-plugin-marketplace/ — CocoLoop — DeepSeek Harness Hits 1.54M Daily Users, 35,000 Plugins: https://news.cocoloop.cn/en/2026/11/deepseek-harness-154m-dau/ — AI Daily Insights — DeepSeek Harness Reaches 1.54 Million Daily Users: https://www.aidailyinsights.cn/en/2026-11-06/8/ — 36Kr — Has the Biggest Fears of Tencent, Alibaba and ByteDance Finally Come True?: https://eu.36kr.com/en/p/4057963796059017

Originally published on AI Growth Lab — https://aigrowthlabshakeel.blogspot.com/