Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Export-focused agent Xiaoyuan AI launches and lands on Tencent's WorkBuddy

Chinese exhibition organiser Meorient unveiled Xiaoyuan AI, an export-focused business agent, on September 23 at the Global Digital Trade Expo in Hangzhou. The product is listed on Tencent's WorkBuddy desktop platform and connects through MCP, letting users run foreign-trade tasks without leaving their workspace.

Published

On September 23, 2026, the fifth Global Digital Trade Expo opened at the Hangzhou Grand Convention and Exhibition Center. In a 14:40-15:00 slot that day, Chinese exhibition organiser Meorient launched Xiaoyuan AI, an AI business agent built for China's export-oriented trading firms, with vice general manager and rotating CEO Deng Meng presenting it on stage.

The pitch is deliberately narrow. Meorient says Xiaoyuan AI is not a generic assistant wearing an export skin, but an agent born inside export business scenarios. It is meant to find buyers, draft outreach letters and negotiate deals, and is described as understanding export work, retaining memory and improving itself over time.

The company frames the capability set as five core tasks, three sources of value and four intelligent loops. According to the launch presentation, those loops connect end to end so that market judgement, buyer profiles, outreach content, communication records and negotiation experience accumulate within the enterprise's authorised boundary and become searchable, callable market, customer, content and method assets.

Commercially, Meorient says the product is lightweight, transparent and consumption-based, a pricing shape intended to lower the barrier for export firms rather than demanding a heavy digital transformation project up front.

Its route to market is the more interesting part. Xiaoyuan AI is listed on Tencent's WorkBuddy office platform, with the two sides connected through MCP, the Model Context Protocol. Users can invoke its foreign-trade capabilities directly inside the WorkBuddy workbench without switching platforms, layering WorkBuddy's general office AI on top of Xiaoyuan AI's export specialisation.

That lands in a suddenly crowded field. ByteDance launched Doubao Work in late August, folding TRAE and Coze into one system built on Feishu's organisational capabilities; Alibaba merged QoderWork, Wukong and MuleRun into Qwen Office in early August, aimed at DingTalk's enterprise market; Tencent consolidated its QClaw team to turn WorkBuddy into a large-scale AI-native desktop office agent.

The export scene is heating up in parallel. Alibaba.com's full-chain agent Accio Work says its AI engine handles more than a billion calls a day, Amazon is rebuilding search and recommendation around Rufus, and Salesforce is pushing Agentforce through sales, service and commerce workflows. Anthropic has published a systemic account of recursive self-improvement, while self-evolving agent research keeps landing across academia and industry.

Note that the Qbitai article is an authorised reprint, so product details come from the vendor's own presentation and should be read that way. The competition it illustrates is real regardless: the focus has shifted from who answers most accurately to who actually gets work done. The signal to watch is whether usage and paid conversion inside WorkBuddy bear out the model, and whether other specialist agents copy this distribution-first route.

Why it matters

Xiaoyuan AI matters less as another vertical agent than as a test of the “platform owns the entry point, specialist owns the depth” split. If it gains real usage inside WorkBuddy, more vertical agents are likely to chase distribution deals rather than build standalone apps.

AgentTencent WorkBuddyAI出海米奥兰特
Back to realtime news

Nearby Updates

All

09/24, 13:45

Lingchu Intelligence Details Psi-R2.5 Data Work: Reversing World Model Psi-W0 to Align Human and Robot Actions

Lingchu Intelligence has updated the data-quality and human-robot alignment methods behind its embodied model Psi-R2.5, building on 100,000-hour-scale data collection. The company says it reverse-uses its world model Psi-W0 to generate matching human-operation data from real robot data, producing strongly paired samples that serve both as training material and as demonstrations for in-context learning.

09/24, 15:00

Anthropic CEO says company will slow down ‘as much as necessary’

Anthropic’s CEO has said publicly that the company will slow its AI development “as much as necessary,” a stance Storyboard18 framed as the AI race meeting its safety brake. The pledge puts safety evaluation ahead of release speed, but the trigger conditions are still defined by the company itself.

09/24, 09:13

Meta Goes All-In on Its AI Agent Muse at Connect

Meta opened its annual Connect event in Menlo Park with a keynote from CEO Mark Zuckerberg that put the AI agent Muse at the center. The company says Muse is coming to Meta's AI glasses as well.

09/24, 08:46

Meta built a Tamagotchi-like wearable for its Muse AI agent

Meta has built a Tamagotchi-like wearable for Muse, its AI agent, according to TechCrunch, giving the assistant another mobile home beyond phones and apps. The device suggests agent makers are now competing on physical form and everyday presence, not just model capability.