China Telecom Launches TeleAgent Desktop AI Office Agent

TeleAgent is designed to work across complete workflows, reading files in different formats, producing editable Excel sheets, PPTs, and web pages, and keeping user context throughout long tasks with a context window of over 400,000 tokens. In APPSO's tests, the agent was put through three realistic use cases: an event plan that repeatedly changed, a student's semester schedule, and a five-person team preparing a product launch.

The first test asked TeleAgent to create an event plan for a simulated coffee brand from folder materials that included budget tables and past case notes. The reviewer then, as a demanding client, changed the event name, cut guests from 160 to 100, cut the budget from 120,000 to 80,000, postponed the start time, and moved to a new venue. TeleAgent updated its working log at each step and delivered a final package that matched every requested change, with no assumptions made about unstated details. The review said this ability comes from the agent's long context, which lets it remember background and follow-up changes.

In the student scenario, course requirements were scattered across syllabi, screenshots, and notifications, with several time conflicts. TeleAgent produced a semester task table, a 16-week plan, and a local webpage that works offline as a weekly dashboard. For a presentation date that had two possible dates from different sources, it kept both and logged the potential conflicts. It also separately arranged around autumn break and exam week. The team scenario involved inconsistent notes, including a quota that changed from 100 to 50 and later to 30. TeleAgent first listed the differences, then generated product positioning, a two-week action plan, a six-page PPT, and a recruit-page prototype. When new changes arrived after the first version, it created a second version while retaining the old one.

APPSO also noted that TeleAgent runs on a credit system that gives new users 3,000 credits, with daily free credits. A Model Router automatically selects the appropriate model according to task complexity, which kept token use lower than expected even for long tasks. The agent lets users limit its workspace, ask for confirmation before key file changes, and keep previous versions; unverified information is left blank or marked pending confirmation. Its skill-sharing platform lets users save and publish reusable workflows. In its conclusion, the review described TeleAgent as a 'new force' and said that cost-effectiveness, delivery quality, and security will decide how many office-agent products become mainstream.