OPPO, Alibaba Close Agent's Organizing Gap on Services and Data

In a separate report on Tuesday, QbitAI detailed Alibaba Qwen Office's open-source project MyContext, which it said had gathered more than 1,000 GitHub stars within a week of launch. The project processes scattered work data—IM chats, documents, meetings, and business records—into dynamic, Agent-consumable context, targeting three enterprise-level pain points: delayed, constantly changing time-series data; contradictory facts; and the computational cost of continuously updating massive historical data.

Leiphone reported that the industry has been moving on multiple fronts to close this gap. Anthropic introduced MCP at the end of 2024 to connect AI applications with external data and tools; Google launched the A2A protocol in 2025 for Agent-to-Agent communication. By April 2026, A2A had support from more than 150 institutions and had entered major cloud platforms. OPPO and Alipay have taken the collaboration further with the AHA (Agent Hub Access) protocol, which breaks through traditional cloud-based A2A limitations and establishes an end-side trusted interconnection paradigm through deep handshake, scenario-based authorization, and brain-hand coordination. OPPO's Xiaobu assistant and Alipay's Abao agent have connected nearly 200 common functions and delivered more than 18 digital public service scenarios, including movie tickets, food delivery, social security and housing fund inquiries, childcare subsidy applications, household registration transfers, visa processing, and utility payments. The integration has contributed millions of daily active users to Alipay, according to Leiphone.

The QbitAI report describes MyContext as a data processing layer for Agent systems. Rather than ordering context purely by timestamp, the project binds each raw message to a stable source identifier, uses conversation-idle intervals as session boundaries, and aggregates evidence in a sliding time window so that repeated facts across discussions raise confidence. For conflicting information, a three-state merge mechanism either reinforces confidence with consistent data, merges supplementary details, or retains multiple facts while lowering confidence and exposing the conflict to users. To control costs, MyContext relies on incremental computation, local rule-based processing for clear-cut cases, batch triggers, version caching, and tiered degradation.

Qwen Office, according to the same QbitAI report, positions MyContext as part of a three-part closed loop with DingTalk and the Qwen Office agent stack. DingTalk, with more than 20 million enterprise organizations and nearly 800 million users, serves as the data entry point; Qwen Office's context engineering governs the data; and the Qwen Office agent layer consumes high-quality context for task execution. The architecture aims to convert the tacit knowledge embedded in chat logs and documents—information that people know but AI does not—into traceable, user- and organization-authorized context that Agents can consume over time, allowing them to participate in real workflows rather than merely respond to isolated prompts.