Published Updated
In one or two sentences, before anything else.
WorkBuddy is Tencent's AI agent for office work, built by the Tencent Cloud CodeBuddy team and launched in China on March 9, 2026. In seven months it added a global edition and phone apps, and Tencent announced a Guangdong government version. It plans tasks, works on your local files and hands back documents, spreadsheets and slides.
What it can do, one sentence each, and whether we actually tested it.
Reads the PDF, Word, Excel and image files in a folder you choose and delivers finished Word, Excel, PowerPoint, PDF or Markdown files.
Tencent's description on workbuddy.cn.
Auto routing or a pick from built-in models including Tencent's Hy4 preview and Hy3, GLM-5.3, Kimi-K3, MiniMax-M3 and DeepSeek-V4-Pro; you can also add your own API key or a local Ollama model.
Model list from WorkBuddy's docs; it may vary by edition and account.
Takes tasks from WeCom, QQ, Feishu and DingTalk in China, and from Slack, Telegram and Discord in the global edition, then runs them on your computer.
Reported by IT之家 (March 9, 2026) and TechNode (May 29, 2026).
Runs recurring tasks hourly, daily, weekly or once at a set time, such as a weekly report.
Tencent's description on workbuddy.cn.
Installs skills, including OpenClaw skills, and connects to outside services through Model Context Protocol servers.
Since version 5.3.5 (July 30, 2026) the AI edits the selected part of a Tencent Docs or local Word, Excel or PowerPoint file while you keep writing in the same file.
Reported by 量子位.
Default permissions pause before risky steps such as deleting protected files, running scripts or going online; full access removes those pauses.
WorkBuddy's docs advise full access only in isolated folders or virtual machines.
Android, iOS and HarmonyOS apps either drive your desktop remotely or run the task in a cloud sandbox.
Launched July 18, 2026, per IT之家.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Graded in
Public tests other people ran: what they gave it and how it did.
Case 1
Input
Build a milk-tea ordering mini-program, then a data dashboard with attribution analysis and suggestions. The same tasks went to 千问办公 and 豆包工作.
What happened
WorkBuddy finished the mini-program in about 6 minutes, against 15 for 千问办公 and 20 for 豆包工作, with mock product images and a best-seller list, the most complete of the three. In the same test its poster was average and its animated SVG had visible flaws.
Case 2
Input
A research-style report with six parts: market growth trend, regional shares, the ten fastest-growing companies, the link between adoption and investment, three industry insights, all in report format.
What happened
WorkBuddy ran for 10 to 15 minutes, crashed several times and still delivered a visual report whose data checked out. The insights and conclusions did not appear in the charts, and the author advises splitting complex tasks into steps.
Case 3
Input
Sort 23 mixed invoices into folders by type, then build an Excel summary with charts.
What happened
WorkBuddy created the folders but put restaurant invoices in the wrong one and extracted only the taxi amounts, so its charts were wrong. DuMate sorted them correctly, with three fields left incomplete. Check WorkBuddy's numbers before you use them.
Case 4
Input
A one-line request for an education product for English teachers, handled by WorkBuddy's MVP expert team, then a change request to deepen the lesson-resource library.
What happened
WorkBuddy asked clarifying questions, wrote the requirements, passed the change to its other agents, made a survey through the 腾讯问卷 connector and published a trial mini-program with login. The article is promotional and the app was not tried on real users.
What it does badly, with the evidence, and what we could not settle.
Weak
Pulls numbers from documents unreliably.
Evidence: In 陈宇明's invoice test (May 11, 2026) it misfiled restaurant invoices and extracted only taxi amounts, so its charts were wrong; DuMate got the sorting right.
Long, complex tasks can stall.
Evidence: In 量子位's March 11, 2026 test a six-part report crashed several times over 10 to 15 minutes, and a web page requested from the phone failed.
Design work is average.
Evidence: AIX财经 (September 20, 2026) called its poster average, its topic ideas generic and its animated SVG flawed, and summed it up as clever but unsteady.
Model training is on by default in China personal editions.
Evidence: WorkBuddy's privacy guide says Help Improve WorkBuddy is enabled by default there, so new conversations may be used for training until you switch it off.
Capacity runs short when a new model launches.
Evidence: On August 31, 2026 Tencent Hunyuan apologised for queues that began on August 28, the day Hy4 preview reached WorkBuddy, and capped each user's free daily use of it (IT之家).
Not for Windows on ARM, or for confidential files on a personal China account.
Evidence: WorkBuddy's FAQ says Windows 11 on ARM64 is not supported. Teams with confidential files should not use a China personal account unless they first turn off model training.
Uncertain
Whether your data is used for training.
Why unsure: The English privacy guide says China personal editions default to allowing training; the Chinese FAQ says server-side data is discarded and never used for training. We could not tell which applies to which account.
How many people use it.
Why unsure: The 20 million monthly and 13 million daily users come from Z Finance via ChainCatcher; Tencent has not published them.
How it scores on a wider task set.
Why unsure: The same July 2026 report cites a 70-question evaluation averaging 2.31 out of 4, with design tasks at 1.45, but names no evaluator or method.
How stable credits and logins are.
Why unsure: One community article says an April 2, 2026 fault wiped some users' credits and blocked logins for about two hours before compensation; we found no second source.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Maker | Tencent Cloud (the CodeBuddy team) | Tencent Cloud is the publisher on workbuddy.cn; Jiemian names the CodeBuddy team. Source: workbuddy.cnchecked |
| Launched in China | March 9, 2026 | Internal testing began February 6, 2026, per Jiemian. Source: ithome.comchecked |
| Global edition | May 29, 2026, at workbuddy.ai | Source: technode.comchecked |
| Desktop | Windows and macOS | The China site also lists Linux. Source: workbuddy.cnchecked |
| Phone apps | Android, iOS and HarmonyOS, since July 18, 2026 | Source: ithome.comchecked |
| China plans (from July 1, 2026) | Free 500 credits a month; Standard 99 yuan a month (70 with auto-renewal), Advanced 199 yuan, Flagship 999 yuan | A 1,000-credit top-up costs 50 yuan; monthly credits do not roll over. Source: workbuddy.cnchecked |
| Global plans | Free 100 credits a month; Pro 10 USD a month or 96 USD a year (2,000 credits a month); Team 40 USD per seat a month | Source: workbuddy.aichecked |
| Built in models | Hy4 preview, Hy3, GLM-5.3, Kimi-K3, MiniMax-M3, DeepSeek-V4-Pro and others; own API keys and local Ollama models supported | Default context window 200K tokens. Source: workbuddy.cnchecked |
| Model training setting | On by default for China personal editions; off by default for overseas and enterprise editions | Source: workbuddy.aichecked |
| Latest version | 5.8.2, released October 16, 2026 | Source: workbuddy.cnchecked |
| Website visits, September 2026 (workbuddy.cn) | 17.7 million (+21.4% on August); 3rd of 15 on AICPB's AI agents list; 9th of 20 on AICPB's AI tools in China list; 2nd of 10 on AICPB's China growth list | Source: mp.weixin.qq.comchecked |
| Website visits, September 2026 (workbuddy.ai) | 1.72 million (+94.2% on August); 13th of 15 on AICPB's AI agents list | Source: mp.weixin.qq.comchecked |
Answer first, then why, an example, the limits and what to do next.
Yes, WorkBuddy is built by Tencent Cloud's CodeBuddy team. Tencent Cloud is the publisher on workbuddy.cn, and when Tencent's PR director Zhang Jun announced the Guangdong government edition on June 16, 2026, IT之家 reported from Tencent that it is built on Tencent's in-house WorkBuddy agent.
WorkBuddy has a free tier, but regular use needs a paid plan. The China edition gives 500 credits a month free and charges 99 yuan a month for Standard (70 yuan with auto-renewal); the global edition gives 100 credits a month free and charges 10 USD a month for Pro.
Limits
Every chat request spends credits, and stronger models and longer tasks spend more.
WorkBuddy runs on Chinese models, including Tencent's own. Its docs list Hy4 preview and Hy3 from Tencent Hunyuan, GLM-5.3, Kimi-K3, MiniMax-M3 and DeepSeek-V4-Pro among others, plus Auto routing; you can also add your own API key or a local Ollama model.
On China personal editions, WorkBuddy's privacy guide says conversations may be used for training unless you switch Help Improve WorkBuddy off. Overseas and enterprise editions default to off. The Chinese FAQ says server-side data is never used for training, so check the setting in your own account.
Next step
Settings, then Help Improve WorkBuddy at the bottom of the panel.
WorkBuddy is faster and more inventive in the public tests we found, but not the most accurate. 豆包工作 won AIX财经's four-job test overall, Baidu's DuMate sorted invoices correctly where WorkBuddy did not, and no public test compares WorkBuddy with 商汤小浣熊, Manus or Coze.
Why
WorkBuddy won the product-manager tasks in AIX财经's test and built a mini-program in 6 minutes against 15 and 20 for its rivals.
WorkBuddy is the office agent and CodeBuddy the coding agent from the same Tencent Cloud team. They share one account and one pool of credits, so a single subscription covers both.
This page is built from Tencent's own site and documentation, news reports from IT之家, Jiemian, TechNode and 量子位, and four public hands-on tests published between March and September 2026. Every number says where it came from, and we checked all of them on October 17, 2026.
According to AICPB's traffic rankings, workbuddy.cn had 17.7 million visits in September 2026, up 21.4% on August: 3rd among AI agents, 9th among AI tools in China and 2nd on the fastest-growing list in China. The international site workbuddy.ai had 1.72 million, up 94.2%: 13th among AI agents.
WorkBuddy went from an internal Tencent tool to a product on desktop, phone and a provincial government platform in about seven months.
| Date | What happened | Source |
|---|---|---|
| February 6, 2026 | Internal testing starts; more than 2,000 Tencent staff use it | Jiemian |
| March 9, 2026 | Public launch in China, with 5,000 free credits on download | IT之家 |
| May 29, 2026 | Global edition, controllable from Slack, Telegram, Discord and WeChat | TechNode |
| June 16, 2026 | A government edition for Guangdong's civil servants is announced | IT之家 |
| July 18, 2026 | Apps for Android, iOS and HarmonyOS | IT之家 |
| July 30, 2026 | Version 5.3.5 lets you co-edit with the AI inside Tencent Docs and local Office files | 量子位 |
| August 28, 2026 | Tencent's Hy4 preview model debuts in WorkBuddy and tasks queue on day one | IT之家 |
| October 16, 2026 | Version 5.8.2, the latest release we found | WorkBuddy |
On usage, Z Finance reported, via ChainCatcher, that three months after launch WorkBuddy had about 20 million monthly users and more than 13 million daily users (iTiger, July 28, 2026). Tencent has not published those figures itself, so treat them as unconfirmed.
The four tests we found agree that WorkBuddy is quick and inventive. They disagree on how reliable it is.
| Edition | Free | Paid |
|---|---|---|
| China, workbuddy.cn | 500 credits a month | Standard 99 yuan a month (70 yuan with auto-renewal), Advanced 199 yuan, Flagship 999 yuan |
| Global, workbuddy.ai | 100 credits a month | Pro 10 USD a month or 96 USD a year; Team 40 USD per seat a month |
Every chat request spends credits, more for stronger models and longer tasks, and monthly credits do not roll over (China pricing, global pricing).
WorkBuddy works inside folders you authorize. Under default permissions it stops to ask before writing to protected paths, deleting important files, running scripts or going online; full access drops those checks for every task on that computer (WorkBuddy). Its English privacy guide says China personal editions have Help Improve WorkBuddy switched on by default, which lets new conversations be used for model training, while overseas and enterprise editions default to off (WorkBuddy). Its Chinese FAQ says server-side data is discarded after use and not used for training (WorkBuddy). The two statements do not match, so check the setting yourself.
No public test we found puts all of them side by side. In the one head-to-head with Baidu's DuMate, DuMate was steadier with data and WorkBuddy better at presentation. 商汤小浣熊 (SenseTime's Raccoon Work) is cheaper to start, with 300 free credits a day and a paid plan from 29.9 yuan a month (SenseTime), but no test we found compares it with WorkBuddy. Manus calls itself an action engine that carries out tasks (Manus), and Coze (扣子) calls itself an AI office assistant platform (Coze); neither has been tested against WorkBuddy in anything we found. WorkBuddy's clearest edge is how it plugs into WeCom, QQ, Tencent Docs and Tencent Meeting.
Sources
Changelog
Runable is a general AI agent that builds websites, apps, slides, documents, images and videos and now also runs ads and outreach, made by a 15-person startup in Bengaluru founded in 2025 by Umesh Kumar and Saksham Sarda. It turned on paid plans in March 2026 and raised a 21 million USD Series A on August 26, 2026.
Polsia is an AI service from Polsia, Inc. that says its agents plan, code and market a company with no human staff. Founder Ben Broca launched it in December 2025 and reports 10,000 paying customers; critics doubt his revenue figures, and his own numbers show nine in ten companies on it have yet to earn a dollar.
Browser Use is an MIT-licensed Python library and a paid cloud service that let AI agents operate real web browsers, made by a Y Combinator startup founded by Magnus Müller and Gregor Žunič. Its GitHub repository had about 118,300 stars on October 17, 2026, up from about 50,000 at its 17 million USD seed round in March 2025.
Raccoon Work (商汤小浣熊) is SenseTime's AI assistant for office work: data analysis, documents and slides on web, desktop and phone. SenseTime merged its office and coding assistants into it on August 25, 2026, and its first-half results report cloud monthly users up nearly tenfold. No independent test of the new version is public yet.
Muse is Meta's personal AI agent: it books, shops, sends email and fills in forms from its own cloud computer, asking before it pays or sends. Meta launched it in the US on September 8, 2026 and has since added Canada. Early reviews are mixed, and Amazon blocks it from buying on its store.
Manus is a general AI agent that researches and builds slides, websites, videos and games, made by a company founded in China in 2022 and based in Singapore. China blocked Meta's purchase of Manus on April 27, 2026; Manus resumed independent operations on September 1 and launched Manus 2.0 on September 28, 2026.