Your Company's Knowledge Base —
AI Customer Service That Answers Right and Meets GB/T 47746—2026

⁠‌‌‌‌‌​​‌​​‌​‌‌‌‌​​​‌‌​‌‌​‌​‌​​‌‌⁠Enterprise knowledge base + AI customer service — built for how small and medium businesses in China actually work.

We turn the documents sleeping in your folders, WeChat groups and cloud drives into a knowledge base that AI can read, search and answer from — with citations — and we configure AI customer service that replies to your customers instantly and hands over to a human when it should. Not a course. A delivered, running system.

On request
deployment pricing
24/7
AI customer service online
GB/T
47746—2026 configured
7×12
support & monitoring

What we do

One knowledge base, two exits: feed your own AI, and be quoted by others
📚

Enterprise AI Knowledge Base

Contracts, quotes, SOPs and product specs are cleaned, deduplicated and made answerable. Employees ask; AI answers in seconds with a citation. Your documents stay on your own machine.

Learn more →
🤖

AI Customer Service

Customer enquiries on WeChat get instant automatic replies grounded in your own product and pricing material. Anything the AI isn't sure about is handed to a human — never improvised.

Learn more →
📐

Compliance with GB/T 47746—2026

China's national standard for human–AI customer service collaboration took effect on 2026-09-01. Our configuration is built around its handover requirements, clause by clause.

See the standard →

Six MCP services — let AI call what we build

Four hosted endpoints (public, read-only, no install, no API key) + two open-source services you run locally. Any MCP-capable agent (Claude, Cursor, Cline, your own) connects directly.
🤖

AI Customer Service MCP

5 tools: clause lookup and standard metadata, the compliance self-check list, atomic answer search and full text.
Endpoint savantcat.cn/mcp — China's AI customer-service standard and enterprise knowledge base, served straight to an agent.

How to connect →
🌐

GEO MCP

7 tools: seven-layer AI-citability audit (0–100), source-pool probe, visibility scoring, weighted fix plan.
Endpoint savantcat.cn/mcp-geo — checks whether AI can find and cite any site.

How to connect →
⚖️

China AI Compliance MCP

8 tools: article-level lookup across four regulations plus mandatory standard GB 45438-2025, scenario self-checks, filing routes.
Endpoint savantcat.cn/mcp-compliance — every conclusion names the provision it comes from.

How to connect →
📡

Content Pattern Radar MCP

3 tools: read a batch of competitor content, surface the claims everyone repeats, the sources the corpus leans on, and how homogeneous it is.
Endpoint savantcat.cn/mcp-radar — turns "what does everyone say" into a verifiable list.

How to connect →
🧭

Content Router (open source, runs locally)

Feed in one piece of material and it profiles it across 11 dimensions, then recommends the formats it best fits (slides, infographic, 60-second video script, comic, newsletter, skill) — each with a fit score and the reasons behind it.
Pure computation, read-only, zero dependencies, no API key.

See how to use it →
🌙

Memory Dream Engine (open source, runs locally)

Lets an AI's memory be tidied the way a person tidies theirs overnight: zero-LLM fact extraction, three-dimensional dynamic scoring, Ebbinghaus 30-day decay, and a five-stage dream cycle.
Pure computation, read-only, no third-party dependencies, no caller data stored.

See how to use it →
MCP overview — connect to all four with one config block →

The standard your AI customer service now answers to

If you run AI customer service in China, this is what “doing it properly” means as of September 2026
NATIONAL STANDARD

GB/T 47746—2026 · 顾客联络服务 人工与智能客户服务协同要求

Customer contact service — Requirements for collaboration between live and intelligent customer services

Published 2026-05-25, effective 2026-09-01. It governs how human agents and intelligent (AI) customer service must cooperate across overall requirements, inbound tasks and outbound tasks.

Clause 5.2.2.6 lists five situations requiring automatic transfer to a human agent — including when the customer explicitly refuses intelligent service, when the interaction involves information security, or when the intelligent response times out (the standard's example: over 3 minutes).

Read the standard explained, with citations →  · Run the 61-point self-check →

Who we serve

Small and medium businesses — the ones who feel every missing answer
🏭
Manufacturing
🩺
Clinics & healthcare
🎓
Education & training
🏪
Retail & trade
💼
Business consulting
🍵
Storefronts on WeChat

Why SavantCat?

Because we deliver systems, not courses.

A course teaches you “how to do it”. We hand over a system that is already running — installed, loaded with your documents, tested against your real questions, and monitored after delivery. What you buy isn't information; it's a working asset plus the method behind it.

Everything is delivered remotely. You don't need to understand technology, and you don't need to keep an IT person on staff.

See products & pricing →

For individuals: the self-service editions

The personal line — the same AI stack, packaged for one person

The knowledge-management method is where our practice started: what we learned building retrieval systems for our own material is what lets us build knowledge bases AI can actually use. The enterprise line — knowledge base deployment plus AI customer service compliance — applies the same thinking to a whole company, which is where our focus is.

以上服务内容、交付范围、时间与效果描述均为一般性说明,具体以双方最终达成的协议约定为准。