A small-business owner in Yichang wasn't sure the contract a buyer pushed across the table was fair. He opened his phone, typed a question, and within seconds a machine answered — in plain language, with the relevant law attached.
It was not a lawyer on call at midnight. It was "Xiaoxiao Si" (猇小司), an AI public-legal-service agent rolled out by the Xiaoting district justice bureau in Hubei province.
Key takeaways
- Xiaoting district (Yichang, Hubei) launched "Xiaoxiao Si" in early 2025; the bureau bills it as the province's first fully self-controlled AI public-legal-service agent.
- The system covers seven core scenarios, from legal consultation and document drafting to business-risk screening.
- According to the district justice bureau, it has logged more than 1,000 resident visits and lifted grassroots workflow efficiency by over 60%.
- It is free to the public and routes complex cases to human lawyers — but it is a first-stop filter, not a courtroom substitute.
What the agent actually does
The name means "little legal aide from Xiaoting." Behind it sits a large model (大模型) wired to a legal knowledge base, designed to behave less like a search box and more like a clerk who has read the rulebook.
Residents can ask in plain text. The agent returns a structured reply: the applicable rule, a short explanation, and often a draft document. For individuals, the headline features are instant legal Q&A and free document generation. For businesses, it offers "scan-to-enter" enforcement-risk assessment and a compliance "health check" that flags where a company might be exposed.
The bureau says the agent draws on multiple models, calling whichever fits the task, rather than depending on a single black box.
Seven jobs, one assistant
The deployment targets seven scenarios across the justice bureau's daily work:
- Psychological profiling for dispute mediation
- Legal consultation for the public
- Legality review of documents and decisions
- Enterprise risk prevention and control
- Compliance "health checks" for businesses
- Smart assessment of "scan-to-enter" enforcement
- General public-legal-service support
The point is coverage. A single township office may see questions spanning family disputes, labor contracts, and business permits in one morning. The agent is meant to absorb the repeatable, well-trodden share of that load.
Why a justice bureau built its own
Grassroots legal offices in China are busy and thin on specialist staff. A villager with a tenancy question, or a shop owner unsure about a clause, often has nowhere cheap to turn quickly. The bureau's pitch is simple: let the machine handle the obvious, let the lawyer handle the hard.
Since it went live in early 2025, the agent has been rolled out across the bureau's divisions, its public legal-service hall, and local justice offices. The bureau reports it has logged more than 1,000 resident visits and raised grassroots work efficiency by over 60%.
For the public, the experience is meant to feel like asking a knowledgeable neighbor — type a question, get an answer in seconds, get a document for free.
Where the human still matters
The system is explicitly a front door, not a replacement. When a case is too tangled for the model — a contested inheritance, a multi-party commercial dispute — the agent hands off to a human lawyer through a one-tap booking. The bureau describes this as "machine screens first, human closes the loop."
That division matters. An AI that drafts a standard lease is useful. An AI that tells a frightened resident their rights in a dispute, in plain words, at 11 p.m., may be more useful still. But neither is a licensed advocate standing next to you in court.
Honest limitations
- It filters, it does not represent. "Xiaoxiao Si" answers questions and drafts documents; it cannot appear in court, give binding legal opinions, or guarantee a favorable outcome.
- Quality depends on the knowledge base. The agent is only as good as the laws, cases, and local rules fed into it. Gaps or stale entries produce confident-but-wrong replies.
- Sensitive data stays sensitive. People describe real disputes to the system. How that data is stored, who can read it, and how long it is kept are governance questions the deployment must answer, not assume.
- Coverage is local. As of this writing the agent serves Xiaoting district in Yichang. Residents elsewhere cannot simply call it up.
- The 60% efficiency figure is the bureau's own. It is a self-reported operational metric, not an independently audited number, and "efficiency" here measures internal workflow speed rather than resident outcomes.
What readers can do now
- If you run a small business in a district that has launched a similar AI legal service, try it for routine contract and compliance questions — then confirm anything high-stakes with a human lawyer.
- Ask your local justice bureau whether a public legal-service agent is available; many are free and worth knowing about before you need one.
- Treat any AI legal answer as a first draft of understanding, never as final advice.