TeleAgent V3.0: AI Agents Now Work on Their Own Windows Cloud PCs

TeleAgent V3.0 Introduces Windows Cloud Computer for AI Agents

China Telecom's Xingchen super intelligent agent TeleAgent has been upgraded to V3.0. The core change is the addition of a cloud execution capability based on a full Windows cloud computer, complementing the existing local desktop execution. This makes TeleAgent one of the first domestic agent products to combine an intelligent agent with a complete Windows cloud environment, better aligning with China's office software ecosystem and user habits compared to Linux-based cloud environments used by some overseas agents.

Cloud Execution: Giving Agents Their Own Digital Workstation

The article argues that for AI agents to become true digital employees, they need their own computer. TeleAgent V3.0 provides an independent Windows cloud system where the agent can install software, execute commands, manage files, and perform flexible desktop operations beyond pre-packaged tools. Users can deploy daily office software and professional tools to build a work environment similar to a real PC.

Test Case: Installing Enterprise WeChat

A simple test instructed TeleAgent to install Enterprise WeChat on the cloud computer. The agent completed the installation automatically without manual control of the cloud desktop.

Test Case: 3D Modeling with Blender

Further testing involved using Blender to model a car. After several rounds of thinking and debugging, TeleAgent delivered a rendered car image and the corresponding Blender project file.

During follow-up attempts to open Blender and inspect the project, the software failed to start. The issue was handed to TeleAgent, which resolved the startup problem in two attempts and restored the previously installed higher Blender version, all without manual system configuration changes.

In cloud execution mode, the agent service runs inside the cloud computer, so task execution does not continuously occupy the user's local CPU resources as long as the cloud computer and services remain running.

Local Execution: Complementary Capability

Local mode retains advantages for tasks requiring direct handling of local files or use of already-installed software. TeleAgent can still call local computer resources within user-authorized scope. A test scenario had TeleAgent operate a browser via GUI to retrieve the day's global AI hotspots and compile a "Daily AI Frontier Intelligence Report" involving multi-source retrieval, news filtering, content summarization, and trend analysis. The agent produced a detailed Chinese report covering multiple AI dynamics with categorized analysis.

Multi-Device Collaboration: Cloud Work, Multi-Endpoint Relay

TeleAgent V3.0 connects desktop, web, and mobile endpoints. All three can initiate and manage cloud tasks. The desktop client handles both local and cloud tasks; the upgraded mobile app manages cloud tasks and can also control desktop TA tasks, enabling management of both execution modes from a mobile device. In practice, task records (including the Blender modeling and debugging sessions) were visible on both web and Android clients, with full execution process and results accessible.

Desktop handles efficient interaction, mobile suits on-the-go task initiation and progress checks, web lowers entry barriers. Devices become entry points rather than determinants of where the agent works.

Memory Synchronization Across Devices

To make cross-device relay seamless, TeleAgent V3.0 introduces memory synchronization. When a user launches tasks from multiple terminals under the same account, a unified memory management mechanism integrates information from different devices. Even with simultaneous web and mobile use, the system periodically merges relevant memories and tags information sources to reduce context fragmentation.

Corresponding memory management and forgetting mechanisms adjust stale information from long-term use, cleaning unnecessary content when needed.

Enhanced Interaction and Platform Support

V3.0 adds real-time speech recognition with filler-word cleanup, shortcut keys, custom skins, and a desktop pet for efficiency and emotional value. New clients include HarmonyOS PC, upgraded Android app, with iOS planned.

Security: Leveraging China Telecom's existing cloud computer security, data transmission uses encrypted channels, cloud computer files are stored encrypted, and multi-replica backup with fault recovery enhances data reliability.

Skill Square, Expert System, and Connector Ecosystem

Beyond a computer, agents need professional skills. TeleAgent V3.0 expands the skill square, expert system, and connectors.

Skill Square

Skills can be searched and installed with one click during task conversations. New skills cover office work, content creation, and personal decision-making: official document formatting per national standards, AI drama planning (script adaptation, character design, storyboard planning, quality inspection), and structured decision frameworks for life advice. Official Office series skills improved mobile adaptation; web mini-app creation skills added data persistence.

Expert System

Users can create and manage dedicated experts by setting name, responsibilities, professional knowledge, and configurations, letting the agent work in ways tailored to specific business scenarios. For example, a marketing planner expert focuses on communication impact and creativity, while an administrative expert emphasizes document standards and format requirements. Pre-configuring experts avoids repeated background explanations per task.

Connectors

New connectors include CamScanner, Canva, Fadada, Jinshuju, QQ Mail, Tencent Docs, Weiyun, Tencent Meeting, Watermark Camera, covering document processing, visual design, e-signature, information collection, email, file management, and online meetings.

Enterprise Features

TeleAgent V3.0 launches an enterprise-exclusive skill square where organizations can publish skills to an internal capability resource pool. An accompanying enterprise management backend supports unified skill management, enabling enterprises to accumulate and reuse business capabilities.

Conclusion: Toward Seamless Human-Agent Collaboration

The article reflects that computer evolution has always aimed to make machines easier for humans to use — from command line to GUI to mobile internet — yet humans still must learn software and perform operations. Agents change this by understanding intent and autonomously executing operation sequences, letting people focus on goals rather than steps. TeleAgent V3.0's upgrades in cloud execution, multi-device collaboration, skill and memory capabilities represent steps toward agents shedding device and tool constraints to integrate more naturally into work and life. True "imperceptible interaction" remains a long-term goal, but human-digital world interaction is entering a new phase.