Tencent opens WorkBuddy Computer Use beta for Windows and macOS apps that lack APIs
Tencent announced a public beta of WorkBuddy's Computer Use feature on November 20. It lets the office agent operate desktop applications that have no API, connector or browser version, on Windows and macOS, and requires WorkBuddy client 5.11.0 or later.
Tencent announced a public beta of WorkBuddy's Computer Use feature on November 20, extending its office agent into desktop applications that offer no API, connector or browser version. Chinese reports describe the feature as software-interface recognition combined with clicking, typing and scrolling. It is not a new model-weight release.
Requirements and setup
- Platforms: Windows and macOS.
- Client version: WorkBuddy 5.11.0 or later.
- Activation: Users must enable the capability themselves and grant operating-system permissions, including accessibility and screen recording.
Tencent cautions that interface recognition varies between applications. Compatibility should not be assumed simply because an application opens on the desktop.
Use cases described
The announcement's examples focus on existing business software. One is looking up records in an education-administration system. Another is matching warehouse and asset information across clients using an incoming-stock reference. The stated aim is to bridge workflows that would otherwise require manual copying between disconnected programs.
Monitoring and limits
A side picture-in-picture window lets users monitor progress while working elsewhere. Simultaneous interaction with the same application is not safe, however: changing its input focus or page state can interrupt the agent's task. This matters most in software where a click depends on the current window, selection or form.
For sensitive actions such as signing in, granting authorization, paying, submitting information or deleting records, Tencent's official guidance is to stay present and check what happens. Users should also return to the original system to verify results once a task completes. These are supervision recommendations, not independently tested guarantees that risky actions will be blocked.
Relationship to Hy4
The feature is a WorkBuddy client-level capability. It is distinct from the vision and computer-use abilities described for Tencent's Hy4 model earlier in November.
November 20 is the date of the public announcement, not necessarily the first date the feature was accessible: an earlier user account described a beta setting. The reports cited are from IT Home, Phoenix Technology and Yicai, as relayed by Pandaily. Pandaily said it did not install WorkBuddy or test the desktop controls. The reports provide examples and operating limits, not a standardized success rate across business applications. Pricing and model details were not part of the announcement.
What this means
Many enterprise tools, especially older internal and sector-specific clients, have no API or browser version, which leaves interface-level control as the only route for an agent. WorkBuddy's beta targets that gap on the two major desktop operating systems.
The operating constraints are significant. Tencent says recognition quality differs by application, and that changing focus or page state can break a running task. Teams evaluating the beta should test each target application individually. They should also keep a person present for sign-ins, payments, submissions and deletions, and verify outcomes in the source system, as Tencent advises. No cross-application success rate has been published.
Related Articles
Tencent Hy team says next model will be substantially larger, with Flash-Lite and 24GB-class options planned
Tencent's Hy team said in a public Reddit AMA that its next model will be substantially larger and that it will keep investing in large-scale reinforcement learning. It also described Flash-Lite, Hy4-mini and models for 24GB GPUs as roadmap items, not released products.
Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license
Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.
Google rolls out computer use in the Gemini app for Mac for multi-step desktop tasks
Google announced on November 19, 2026 that computer use is rolling out in the Gemini app for Mac. Gemini can carry out multi-step work across local files, desktop applications and websites, working in an isolated background window with a live preview in the chat.
AWS Quick adds chat-built flows, multi-agent orchestration and an updated web app
AWS Quick users can now describe a business process in chat and have Quick build a flow that orchestrates agents, skills, tools and data. The release also adds multi-agent orchestration, an updated web app with a daily briefing, Meta ads and Amplitude connectors, and new language support.
Comments
Loading...