Tencent, long known for a measured and conservative approach to product rollouts, has recently taken a more energetic turn. The company recently generated public attention for OpenClaw through an offline event offering free installations, while also releasing several related products in quick succession.
Among the tools already available are WorkBuddy, an office assistant sometimes described as Tencent’s version of a “deployment free OpenClaw”; QClaw, which supports one-click startup and integrates agentic artificial intelligence capabilities into both WeChat and QQ; Lighthouse, Tencent Cloud’s lightweight application server designed for developers; and a surrounding ecosystem of tools that includes SkillHub and several security solutions.
Tencent CEO Pony Ma revealed on WeChat that the company is also developing a broader lineup.
OpenClaw is an open-source agent framework that can be deployed directly on a local computer, allowing it to operate systems, access files, and respond to messages in ways that resemble human workflows. Unlike a human operator, however, the agent can run tasks continuously without interruption.
Since the Lunar New Year holiday, OpenClaw has become a new arena of competition among China’s major technology companies. ByteDance’s cloud platform Volcano Engine launched ArkClaw, while Alibaba Cloud’s Tongyi Lab introduced CoPaw. Both initiatives aim to lower the barrier for users to install and experiment with OpenClaw-style agent systems.
Despite the surge of interest, questions remain about the technology’s practical value. What can it realistically accomplish in everyday scenarios? And how can users avoid risks such as information leaks or accidental file deletion?
On the evening of March 12, Tencent’s taskforce for OpenClaw-related initiatives spoke with several media outlets, including 36Kr. Participants included Ding Ning, head of developer AI products at Tencent Cloud; Zhong Yucheng, director of Tencent’s lightweight cloud business; Su Jiandong, general manager of cloud security at Tencent; and Xie Yizhi, head of Tencent’s AI agent security center.
They discussed how the installation event came together and addressed questions about security, commercialization, and Tencent’s growing ecosystem of OpenClaw-like products.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: How did the March 6 event outside Tencent’s headquarters come about? How was it prepared?
Zhong Yucheng (ZY): The whole process happened very quickly. On March 1, we distributed several hundred free OpenClaw trial vouchers internally at Tencent Cloud. We thought that would be more than enough, but they were claimed in less than 20 seconds.
After that, many colleagues began asking how to install and run the system. So we thought, why not hold an offline installation session at the north plaza of Tencent Seafront Towers?
At first, the idea was quite simple and intended only for internal staff, so we scheduled it for Friday, March 6. But when we mentioned the idea to others, they suggested opening it up to people outside the company who were interested in AI.
Two days before the event, we posted a teaser video on WeChat Channels. Unexpectedly, the video went viral and far exceeded our expectations.
Initially we recruited only ten volunteers and gave them one or two hours of quick training. They were taught how to help users deploy the system and demonstrate basic scenarios. After seeing the video’s popularity, we expanded the volunteer team to twenty.
On the day of the event, users began lining up as early as 9 a.m, so we temporarily brought in nearly ten additional volunteers from the cloud server computing team.
The crowd was notable. People from many walks of life showed up: a retired engineer in his 60s, a mother pushing a stroller, and even elementary school students. About 80% of participants had no technical background and were simply curious about AI.
We had originally planned to spend about five minutes installing the system for each person. In reality, explaining and debugging it one-on-one took 20–30 minutes on average, and some participants stayed the entire day. In total, we helped more than 500 people complete installations on site.
The level of attention the event generated was far beyond what we expected.
36Kr: So it was really just a spontaneous activity? Was there any intention to use the event as marketing for Tencent’s other products? How did the company evaluate it afterward?
ZY: Honestly, we didn’t think that far ahead at the beginning. At the time, WorkBuddy and QClaw were still largely internal products and had not been formally released.
Our initial motivation was simple. Deploying OpenClaw in the cloud is more complicated than installing it locally, and it requires some explanation for users. Once the event gained traction, different product teams quickly coordinated with each other. That collaboration eventually formed what is now Tencent’s broader OpenClaw product lineup.
36Kr: Many developers say the technology consumes significant computing power and memory. Has Tencent Cloud seen a noticeable increase in token consumption over the past month? Could sustained popularity put pressure on computing resources?
ZY: Because token statistics involve user privacy and permissions, we don’t track them in detail. But it’s fair to say that overall model computing consumption has increased compared with before. We just don’t have precise numbers yet.
36Kr: On X, someone once tricked OpenClaw into revealing an API key through a prompt injection. If I use WorkBuddy and it encounters a similar prompt, could it also leak my private API key?
Ding Ning (DN): It’s important to clarify the difference here. OpenClaw is open source, while WorkBuddy is a closed-source product.
From an architectural perspective, WorkBuddy integrates capabilities we have developed and deployed for years, such as systems used in CodeBuddy and Cloud Studio, including Gateway Server and Agent OS.
We also built a mature sandbox mechanism for developers several years ago. With some adjustments, it becomes the foundation for Agent OS.
Combined with solutions from Tencent’s security lab, we have strong defenses against issues like software supply chain poisoning.
The most important point is that WorkBuddy only operates within user-designated folders. For example, if you ask it to organize or convert files locally, it works only within the specified directory. Its permissions are limited, so the kind of cross-boundary behavior you’re worried about should not occur.
36Kr: What if someone pulls malicious code from GitHub or attempts to poison SkillHub through WorkBuddy? Can you detect that?
Xie Yizhi (XY): Yes. Tencent has multiple internal labs, and we’ve introduced an automated review system based on “agent-versus-agent” defense.
There are several layers of protection ranging from traditional rule-based and signature-based detection to AI-driven systems designed to identify malicious code. These mechanisms work together to prevent harmful plugins from entering the ecosystem.
This capability is integrated into both the creation and upload stages. Once a risk is detected, the system will automatically block it.
36Kr: Some users worry that WorkBuddy could bundle pop-up ads or other unwanted features.
DN: The WorkBuddy team is the same team behind CodeBuddy, which focuses on AI coding. Our background is in developer ops, code repositories, and artifact repositories. Advertising simply isn’t part of our DNA.
The capabilities built into WorkBuddy are all reviewed by our team. Problems are unlikely there. But if users install plugins from third parties, that’s similar to installing unverified apps on a phone.
36Kr: The popularity of OpenClaw reminds people of the moment when ChatGPT launched. What’s different this time? How do you see the relationship between large models and OpenClaw?
DN: This represents a shift from conversation to execution.
Large models initially appeared as chat interfaces. OpenClaw accelerated the spread of a different model of interaction.
As AI coding moves toward autonomous development, backend agents and skills can support broader productivity scenarios. Even people who don’t know how to code can benefit from what feels like a “say it and it happens” experience.
OpenClaw didn’t invent this concept, but its popularity has made the idea much clearer to the public.
ZY: I don’t think OpenClaw represents a breakthrough on the scale of ChatGPT. Instead, it allows ordinary people to experience AI agents more directly.
Many of the things it can do were technically possible with tools in 2025. But the barrier to entry was too high. Even though configuration is still somewhat complex, OpenClaw is currently one of the easiest ways for most people to experience how AI can change everyday life.
36Kr: Some critics say OpenClaw’s popularity exceeds its actual capabilities. Others have suggested it was created mainly to ease anxiety around slow token consumption. What do you think of these views?
ZY: Some level of anxiety probably exists. Right now, it can’t yet deliver the “one-person company” vision people talk about.
But the offline event we held at Tencent’s headquarters tells a different story. Among the participants were a retired engineer in his 60s and a mother pushing a stroller.
Some people online mocked them for wasting time because they didn’t understand how the technology works. But in my view, simply taking the step to experiment with AI already puts them ahead of most people.
For many, OpenClaw provides a tangible entry point into the AI wave.
36Kr: WorkBuddy was already used internally in January. Why launch it publicly now? Was it to ride the wave of attention?
DN: There is a connection. The attention accelerated our schedule.
Originally we planned to release it in mid-March. Last year, when DeepSeek became popular, many people were still on holiday. This year we prepared earlier to avoid that situation.
Since the timing after the Lunar New Year felt right, we moved forward with the release.
Compared with the January internal version, the biggest upgrade now is integration with instant messaging tools through persistent connections. Users can control it remotely through mobile QQ, Feishu (Lark), or WeChat.
If we want to build AI teams in the future, these scenarios need to be supported early.
36Kr: Tencent now has multiple OpenClaw solutions including WorkBuddy and QClaw. Will these eventually merge?
DN: When high-value business scenarios require integration, we will naturally connect them.
Right now we are still in an early stage. Our priority is to build products around successful, closed-loop use cases rather than imagining a product form in advance.
ZY: WorkBuddy is currently more of a local version of OpenClaw, while Lighthouse provides a cloud-based alternative.
On the cloud side, we aim to offer a simpler OpenClaw deployment experience. In the future, we may explore ways for local and cloud versions to interact or merge.
36Kr: Pony Ma has said Tencent strongly supports internal incubation of AI applications. What kind of resource support do these teams receive?
ZY: For the lightweight cloud business, the biggest support comes from computing resources.
Teams behind Tencent Docs, Tencent Maps, and the Emma knowledge base also provide a growing ecosystem of skills that expand potential use cases.
We also launched SkillHub, a skill marketplace designed to be safer and more accessible for Chinese users. It addresses issues such as overseas access failures and compliance filtering.
Different teams operate differently, but overall everyone is moving quickly within their respective domains.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Wang Yuchan for 36Kr.