The trial version can connect to Xiaomi devices and workplace tools, set conditional routines and complete tasks such as generating and printing a home equipment report.
Ant Group’s Lingguang app is testing a Personal Agent feature with a limited group of users, marking a clear shift from a general assistant towards something closer to a household operating layer. In the new version, a “My life” tab appears at the bottom of the app and lets users assign everyday responsibilities to the agent, from checking whether doors are locked and appliances are off to handling device-linked routines without further prompting. The concept is ambitious: rather than answering isolated queries, the assistant is being positioned to monitor routines, act on connected services and intervene only when something needs attention.
The most striking part of the trial is how widely the agent can reach into other systems. According to the user report, it can connect to Xiaomi’s smart-home ecosystem through Mijia, gaining permission to manage devices such as locks, lights and washing machines. It can also link with workplace tools such as DingTalk, suggesting that the product is being shaped as a bridge between home automation and digital coordination. That breadth matters because many consumer AI products still stop at recommendations or reminders; here, the agent is being used to execute tasks across external platforms.
The setup is also unusually hands-on. To connect Mijia, the user simply entered a request to install the Mijia API. The assistant then found the open-source mijiaAPI library, installed it inside a dedicated Python environment and generated a QR code for pairing. Once connected, it detected 13 devices and grouped them by room, including a Xiaomi laser printer. The report says the agent was able to use that link to produce an A4 Word document listing the home’s devices, their online status and their capabilities, then output it on the printer after converting the file through PDF and image formats where necessary.
That workflow is important because it shows the assistant doing more than simple automation. It searched for a code library, set up its own cloud-based Python environment and selected an output method that matched the device’s capabilities. Where a conventional assistant might have stopped at drafting the content, Lingguang appears to have handled the full chain from retrieval to formatting to printing. The result is closer to a task-performing agent than a chat interface, and it hints at why the product is being described as giving an unusually strong “AGI-like” impression.
The same pattern extends to routine automation and shopping research. The user described asking the agent to watch the Beijing temperature each morning and switch on a workroom air conditioner if conditions fell below a chosen threshold. Another example involved tracking discounted robot vacuum cleaners across JD.com and Tmall ahead of Singles’ Day, with the agent monitoring prices and issuing reminders when a target was reached. A further test asked it to watch round-trip flights between Beijing and Chengdu and only notify the user if fares dropped under a set limit. These scenarios are not groundbreaking individually, but together they show an agent moving towards continuous, conditional decision-making rather than one-off replies.
For now, the feature remains in a small-scale test phase, and the report treats the experience as a preview rather than a final product review. Even so, the wider direction is clear. Ant’s Lingguang is being developed as a multi-environment personal agent that can link to connected devices, office software and web services, echoing the broader industry shift towards cross-platform assistants. Open-source projects such as nanoMuse and framework proposals like XuanJi point in the same direction, while tools such as tunnelbox show growing demand for agents that can be supervised from a phone and used safely across local and cloud-connected devices. Lingguang’s Personal Agent suggests Ant wants a share of that market with a product designed to act, not just answer.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
