Ant Group has released LingGuang 2.0, an update that turns its multimodal AI assistant into a more proactive personal agent.
The original app launched in November 2025 and picked up over 2 million downloads in its first six days. The new version is now rolling out in China on iOS, Android, and the LingGuang website.
Most AI assistants tend to sit idle until they receive clear instructions. However, LingGuang 2.0 aims to take action sooner. With the user’s permission, it can check all connected accounts and services, assess issues that might be of concern, start working on them, and alert the user only when something requires their input.
To make this work, Ant Group improved the system’s ability to remember context across different sessions, schedule tasks that stretch over time and call external tools. Because the agent runs on its cloud instances, it can keep going after a chat ends and pick the task back up later if something changes.
Ant Group is already showing a few simple examples of how this could work day to day. LingGuang will track user spending, identify automated transactions, assess payment patterns, and suggest ways to save money once users authorize access through Alipay.
Once users book flights or hotel rooms, LingGuang will track these bookings in real time to keep an eye on price changes. It will keep track of significant dates such as birthdays or car maintenance appointments and will notify users of any events needing attention, provided the users give consent. The idea is to eliminate the need for people to do everything by themselves.
For a system to perform numerous tasks continually, it requires more than a powerful machine learning algorithm. The system must be capable of acquiring information from various applications, using various software, and staying online reliably. LingGuang 2.0 is already linked to Alipay services and can call tools such as MCP and CLI outside that ecosystem.
Presently it can handle Alipay payments, memberships, and travel bookings, as well as email, DingTalk, Lark and Mi Home smart devices. Over time the company says the agent’s reach will grow from personal information and online tasks toward workflows that cross different platforms and control smart home devices.
Questions around permissions, data security, and how much trust users can place in an agent are handled through APASS, Ant Group’s system for managing accounts and security across agents, devices, apps, and services.
APASS builds identity checks, intent verification, and privacy protection into the process of carrying out tasks. Authorization keys are kept separate from the agent and held in escrow, so the system can act without directly holding the user’s credentials. People can review, change, or cancel permissions whenever they want.
A similar move toward proactive agents is happening outside China. Anthropic recently introduced a research preview of Claude Duet, an app in which the AI also works as a proactive agent. Duet runs on the Claude Opus 5.5 model in a cloud-based virtual machine, allowing it to continue tasks even when the user’s phone or computer is turned off. It is aimed primarily at small business owners and professionals who need help prioritizing multiple daily tasks.
The agent does not perform important or sensitive actions on its own. It requests confirmation before sending messages, signing documents, or making payments. Financial actions such as securities trading and money transfers are blocked at the system level.
Each morning Duet provides a short summary of significant updates from connected apps. During the day it monitors those apps and alerts the user when something requires attention. Users can also assign specific tasks such as scheduling meetings, building dashboards, preparing financial or sales forecasts, creating prototypes, and developing web apps.
Both examples show AI assistant developers moving from a “respond to a request” model toward one in which the agent manages tasks on the user’s behalf.
