E-House has opened a public beta in Zhuhai for E-House Xiaoxin, an AI-assisted service intended to guide property buyers and sellers through parts of the home-sales process. The test is notable because it places an AI agent inside a transaction category where the practical work extends beyond answering questions. A property sale can involve listings, pricing discussions, viewings, legal documents, inspections, and a chain of decisions that are difficult to compress into a simple chat interface.
E-House Xiaoxin is the product name used by E-House (China) Enterprise Holdings Limited, a property-services group. Reports describe the Zhuhai beta as offering zero-commission intermediary services for both buyers and sellers through December 31, 2026. The company is not claiming to remove people from the process entirely. Instead, it is pairing AI agents with human “house companions” and third-party fulfillment partners.
That hybrid design is the central fact of the launch. It acknowledges that AI can organize information and guide a user through standardized steps, while property transactions still require offline work and professional accountability. The test will show whether that division of labor is useful to consumers, rather than merely attractive as a cost-saving claim.
A public beta with more than 500 Zhuhai listings
At the time of the beta launch, reporting cited more than 500 verified housing sources on the platform, spanning second-hand homes and new property projects in Zhuhai. Leju Finance and Fangchan.com provided additional coverage of the launch and the local service model.
The number is useful because it defines the initial scope. E-House Xiaoxin is not a nationwide public product with a verified inventory across every Chinese city. It is a city-level beta that needs enough listings to let users test search, matching, and transaction support in a real local market. The 500-plus figure is company or launch-report information, not an independently audited measure of available properties, and should be understood as such.
DeepLinkRE-LLM targets a vertical property workflow
E-House says the agent is powered by DeepLinkRE-LLM, a vertical large language model trained on CRIC real-estate data. The claim comes from the company’s launch materials and should be treated as a description of its system, not as an independent assessment of model quality.
A vertical model is potentially useful in real estate because users ask questions that depend on local listings, property categories, transaction conventions, and market terminology. A general-purpose model may produce fluent language without knowing whether a listing is current or what documentation a specific transaction requires. E-House’s strategy is to combine its data resources with a product designed for a defined consumer workflow.
That data advantage will be meaningful only if it remains accurate and is paired with clear human oversight. Property information changes quickly. A model can summarize historical data, but a user needs to know whether a particular listing is still available, whether price information is current, and who is responsible if the information is incomplete. The beta is an opportunity for E-House to demonstrate those operational safeguards in a single city before making wider claims.
The company’s inclusion of human companions and outside service providers is therefore more than a marketing detail. Reports describe third-party fulfillment partners in the service network. The structure recognizes that an AI agent can coordinate a path through a transaction, while a legal review or physical inspection requires accountable specialists. It is closer to an AI-assisted service operation than to a fully autonomous broker.
China’s agent market has already produced examples of AI supporting discrete professional tasks. EastFrontier reported that a Guizhou AI agent reduced small-business credit-check time. The Zhuhai property beta applies the same broad idea to a more consumer-facing process, where recommendations and information quality can affect a household’s large financial decision.
Property transactions expose limits that can be obscured in a simple AI demonstration. The user may ask for a recommendation, but the answer depends on real inventory, legal status, financing, personal circumstances, and an in-person assessment of a home. The service also has to deal with the trust issue created when an agent is operated by a company with data, product, and commercial incentives of its own.
E-House’s use of a zero-commission beta can be seen as an attempt to address part of that trust issue. It gives consumers a reason to try the agent without paying a conventional intermediary fee. But the ultimate test will be whether users believe the system’s property information is complete and whether the human support layer is available when an automated flow reaches a complex point.
The beta is also a reminder that agent adoption does not always mean removing people. In the E-House model, agents and human companions occupy different parts of the workflow. The AI can potentially handle discovery, first-pass matching, and repeat questions. Humans can handle visits, inspections, legal coordination, and cases where the system needs escalation.
An earlier EastFrontier analysis of AI agent task completion underscores the importance of judging agents by whether they complete real tasks, not just whether they produce plausible language. E-House Xiaoxin faces that same standard. A polished property conversation is not enough if the service cannot help a buyer move accurately from a search query to a verified listing and then to the appropriate offline support.
A local pilot before any national claim
E-House Xiaoxin’s Zhuhai beta is best understood as a limited deployment. The company has not published an expansion timetable for other cities, an independent customer-satisfaction measure, or a verified transaction-completion rate. No publicly disclosed funding allocation for the DeepLinkRE-LLM program has been reported.
Those omissions matter because property technology products often look scalable before they encounter local inventory systems, different regulations, and high-touch consumer expectations. A city-level beta lets E-House examine whether its AI agent can manage the transition between online information and offline fulfillment without promising national coverage prematurely.
The strongest argument for the product is not that an AI agent can replace a broker. It is that a vertical system with property data, a defined user flow, and human service partners may make parts of a complex transaction clearer and less expensive. The zero-commission period gives the company a way to test that proposition with real users.
For now, E-House Xiaoxin is a Zhuhai experiment with more than 500 reported listings, a defined beta period, and a hybrid service design. The next evidence will come from use: whether buyers and sellers return, whether human companions can resolve the cases that software cannot, and whether the company can show that AI assistance improves a property process without reducing the safeguards that make a major transaction trustworthy.