Ant Group launches Lingguang 2.0 AI assistant for long-running tasks

The central change is where the AI enters a task. Previously, users had to spot a problem and ask the assistant to act. With user authorization, Lingguang 2.0 can use connected information and services to identify potential needs, start analysis, monitoring and follow-up, and alert users when attention is required. The upgrade is built on cross-session memory, long-range task scheduling and tool connections. An independent cloud host allows tasks to keep running after the user leaves a conversation, with the assistant following new information and changes.

The report described several authorized scenarios. In consumption, after connecting to Alipay, Lingguang can review bill records, find auto-renewal subscriptions that may have been forgotten, analyze spending habits and offer money-saving suggestions. In travel, after a flight or hotel booking is completed, the assistant can keep watching for price changes and notify the user when a lower price appears. In family use, it can follow teacher notices in a DingTalk class group, summarize important information, extract to-dos and complete follow-up work such as document formatting.

Lingguang can also handle matters that require long-term memory and follow-up, including pet care, child vaccinations, family anniversaries and vehicle inspections. Within the scope of authorization, it can identify key dates, arrange follow-ups and reduce the burden of repeated checks and reminders. Leiphone reported that this marks a shift for AI agents from task executors to affairs managers.

Such work requires more than model understanding and reasoning. It also needs access to information across applications, tool invocation and infrastructure that can run for long periods. Lingguang 2.0 connects data and services in the Alipay ecosystem and supports tools such as MCP and CLI in open environments, according to the report. It has built-in access to Alipay bills, membership and travel services, as well as DingTalk, Feishu, email and Mijia. Its management scope is expanding from personal information and online matters to cross-platform workflows and smart-home device control.

The expansion brings new challenges around permission boundaries, data security and execution trust. Lingguang 2.0 uses Ant APASS, described as unified account and security infrastructure for cross-agent, cross-device and cross-application scenarios. APASS incorporates identity authentication, intent security and privacy protection into agent execution. Authorization keys are separated from the agent and held under separate management, so the agent can perform tasks without directly touching the keys. Users can view, adjust or revoke authorization at any time.

Lingguang App, Ant's multimodal general-purpose AI assistant, was launched on Nov. 18, 2025, and recorded more than 2 million downloads in six days, according to the report. Earlier features included one-sentence generation of flash apps and a function called Generate Your World, extending natural-language interaction to app generation and multimodal content creation. The report said the next stage of competition for AI assistants will depend less on how many questions they can answer or commands they can complete, and more on whether they can understand long-term needs, intervene at the right time and keep managing matters. A chatbot mainly solves immediate problems, while a personal agent must handle matters spanning hours or days, remember context, sense changes, advance tasks and ask for human judgment when necessary.