SavantCat — Enterprise Knowledge Base & AI Customer Service

Who we are

SavantCat (合尘猫) is a one-person AI studio based in China. We build enterprise AI knowledge bases and deploy AI customer service for small and medium businesses — the companies that feel every unanswered question and cannot justify an IT department.

Our clients are SMEs under 200 employees: chain stores, professional-service firms, small manufacturers. Their knowledge is concentrated in a handful of people, a folder of contracts and a few old spreadsheets. That is exactly the material a knowledge base has to survive.

We publish our own fact sheet so you can check who we are, what we deliver, how pricing works and how the claims on this site can be verified — before you talk to us. Where something has not been verified, the page says pending rather than filling the blank with a number.

What we do not sell is courses. What we hand over is a system that is already installed, loaded with your material, tested against your real questions and monitored after delivery.

What we do — three things

• Enterprise knowledge base building. Your product data, policy clauses, service rules, contracts and quotes — currently scattered — are turned into a knowledge base that answers with citations: inventory and de-duplication, structure design, chunking rules, an evaluation set built from your own questions, then post-launch checks. Deployed on your own machine, not in someone else's account.
• AI customer service and national-standard compliance. Replies grounded in your material, with handover to a human when the AI is not confident — configured against GB/T 47746—2026, China's national standard on collaboration between live and intelligent customer service, effective 2026-09-01. We self-check all 61 items (48 shall + 4 should + 9 may), including the 5 veto items, and hand over the sheet with clause references.
• Answer-layer building (AEO, also called GEO). The same verified material published as structured public pages and machine-readable files, so search engines and AI answer engines can find the business, read it and cite it. We also run two public MCP servers any AI agent can call — see MCP endpoints.

→ See how the knowledge base is built · → AI customer service and compliance · → Run the free 61-item self-check

The methodology behind it: what enterprise knowledge engineering is

Enterprise knowledge engineering is not a tool you buy. It is a system that makes knowledge flow inside a company. The point is not storage — it is movement: from collected once and forgotten, to collected, organised, connected and used.

We all collect constantly: an article saved to a chat app, a video saved to a phone, a report downloaded to a drive. Then nothing happens. A knowledge base only earns its keep when it does three things:

• Findable. Retrieval works by meaning, not by folder name. You ask in ordinary language — "which clause covers response timeouts?" — and the system surfaces the clauses and the answer, even when the sources live in different files.
• Reusable. One cleaned document serves customer service this month, a quotation next month and a training note the month after. Knowledge stops being a one-off purchase and starts compounding.
• Growing. Every question the system cannot answer is a visible gap: it tells you which document is missing or which version contradicts another. That is how a knowledge base gets better instead of quietly ageing.

The same material then feeds two exits: inward to your own AI customer service and staff, outward to the AI answer engines your customers already ask. One source of truth, two jobs.

Why we built this

I am the person this started with. Over the years I collected more than ten thousand articles, downloaded hundreds of gigabytes of material, and genuinely used less than 5% of it. Whenever I finally needed one fact, either I could not find it, or I had forgotten why I had saved it in the first place.

The second problem was cost. From 2024 onward, AI tools kept getting more expensive: monthly subscriptions multiplied, prices rose, and the material stayed in somebody else's account. As an independent practitioner, both trends pointed the same way — I wanted the knowledge management problem solved once, the payment made once, and the data on a machine I control.

SavantCat is the answer I then packaged for companies: not a subscription, not a course, but a delivered system. Your documents are cleaned once, the answers carry sources, and the whole thing keeps working after the invoice is paid — because you own it.

Services, and how pricing works

• Enterprise AI knowledge base — one-time deployment plus an annual service term: your contracts, quotes, SOPs, product data and policies become searchable, and staff get cited answers in seconds. Light private deployment; monthly inspection, content updates and remote operations are included.
• AI customer service system — sized to your consultation volume: instant replies on WeCom and web, grounded in your product and price material, with automatic handover to a human whenever the AI is unsure. Where the standard applies, the compliance self-check and remediation are done in the same project.
• Answer-layer and open-data work — the checklist data, fact pages and MCP endpoints that make your material citable, so AI assistants can quote a verifiable source rather than guess.

Pricing is quoted per project — email savant0196@126.com. We inventory your documents first, turn them into a problem list and a plan, and only then quote. Nothing on this site is published as a fixed price, because a knowledge base project's price depends on volume, structure and integrations.

A project runs in six steps, each with a written output and an acceptance rule, and three gates that stop the work rather than push it forward — read the delivery process. The full service list and what counts as delivered are on Services & delivery standards.

What we do not promise

• Search ranking positions.
• Citation rates, inclusion rates or recommendation-rate numbers from any AI platform.
• Blanket coverage of AI answers about your business.
• Audit conclusions, certificates or compliance certification we are not entitled to issue. GB/T 47746—2026 is a recommended national standard, and the third-party assessment described in its clause 9.2 must come from a body holding professional customer-service qualifications.

Those first three are black-box outcomes of third-party platforms, and no service provider can honour them in a verifiable way; the applicable standard, T/CAPT 026—2026, explicitly prohibits service providers from promising such outcomes. What we do commit to is a verifiable process: what was done, how, and where the raw results are — you can re-run every step yourself.

Principles we work by

• Knowledge has to move. Saving a file is not knowing something. A system that can be searched, reused and grown is worth more than a bigger archive.
• AI should assist thinking, not replace it. The AI drafts, retrieves and reminds; a person still owns the answer that leaves the company.
• Data ownership is not a slogan. Your documents stay on your side and your material is never sent to overseas APIs. Privacy, cost control and the freedom to customise all follow from that.
• Knowledge compounds. Every cleaned document and every closed gap makes the next question cheaper to answer. That is the return we sell.

Frequently asked

Do you publish prices? No. Pricing is quoted per project — email savant0196@126.com. A public number would be meaningless for your case, and we would rather spend the first call on your documents than on a price list.

What do you need from us to start? Where the documents live (contracts, quotations, SOPs, product material, support logs) and one person who can confirm which version is authoritative. That is the whole prerequisite.

Can our own staff maintain it afterwards? Yes — that is the acceptance standard. A project is not delivered until one of your own people adds a document and watches it take effect.

Contact

• 📧 Consulting & purchase: savant0196@126.com
• 🧭 Enterprise knowledge base: see how it is built
• 🤖 AI customer service: deployment and standard compliance
• ✅ Free self-check: GB/T 47746—2026, 61 items
• 📄 Verify us first: fact sheet · privacy and data
• 🔌 Machine-readable: MCP endpoints — public, read-only, no API key
• 📱 Where else we publish: CSDN · Juejin · Weibo · Zhihu · WeChat Official Account — all under the name SavantCat (合尘猫). Account links are listed on our fact sheet; this English page carries no off-site links.