AI Website/AI Customer Service/Enterprise knowledge base/Data Workbench/Expiry alerts/Executive AI Copilot/FDE On-site Delivery/Coaching & company-wide training/
AI Website/AI Customer Service/Enterprise knowledge base/Data Workbench/Expiry alerts/Executive AI Copilot/FDE On-site Delivery/Coaching & company-wide training/
Agent · Enterprise AI Agents — 4 standard products + delivery services
Before AI can work for you,it has to {{heroWord}}.
Website, customer service, data dashboard and executive assistant — four products sharing one company knowledge base. No need to launch everything at once: pick the most painful use case, get results, then expand.
know the businessget work doneknow its limits
magic.stack · product architecture
Illustration
{{t.no}}{{t.name}}
Enterprise knowledge baseOne knowledge base, reused everywhere
Policies & processesProduct informationBusiness SOPsPast Q&AThe boss’s judgment calls
Connects to: ERP · inventory · Excel · WeCom · Feishu · website forms
{{arch.caption}}
Enterprise AI Agents · four standard products
Use one on its own or combine them; each new product reuses the existing knowledge base
PRODUCT 01 · Built for people and for AI
AI Website
AI Website
Built-in assistant
A website that greets visitors and captures leads. Pages are structured so AI can read them, Magic Assistant answers visitors’ questions, and contact details go straight to sales.
— Website design and build, great on desktop and mobile
— Products, certifications, cases and FAQs as structured pages
— Magic Assistant greets visitors, with answers drawn from your knowledge base
— Leads pushed automatically to WeCom or Feishu
Best for: Companies whose website hasn’t been updated in years, who answer inquiries by hand, or who are getting ready for GEO
Pricing: Quoted by number of pages and features, including the assistant
How GEO works →
www.[your-company].com
[Company]Products · Solutions · Cases · Contact
[What you do, in one sentence]
Product specs
StructuredCertifications
VerifiableCustomer cases
Sourced
StructuredCertifications
VerifiableCustomer cases
Sourced
Magic Assistant
Hi, which product are you interested in?
Is [model] in stock?
Yes, in stock — ships within [X] days. Leave a phone number and sales will contact you right away.
New lead · sent to WeCom
PRODUCT 02 · Customer service and knowledge base, as one
AI Customer Service
AI Customer Service
+ Enterprise knowledge base
First we turn your company materials into a knowledge base AI can use, then customer service answers from it. Replies to customers and answers for staff come from the same knowledge.
— Not just policy files: product information, SOPs and past Q&A organized together
— Every answer cites its source; when unsure, it hands off to a human
— Which questions AI answers directly and which need a human are written down before launch
— New questions are collected and added to the knowledge base, so it keeps getting better
WebsiteWeComWeChatFeishuInternal staff assistant
Tip: Start with pre-sales inquiries. After-sales issues like refunds and complaints are more complex and are assessed separately.
Pricing: Setup fee (materials, knowledge base build, launch training) + annual service fee — quote on request
Customer chat · WeComIllustration
[X]-year warranty on the whole unit, [Y] months on wear parts.Source: “After-sales Service Policy”, clause [N]
Pricing needs confirmation from sales — I’ve passed this to [sales rep], who will contact you shortly.Handed to a human by rule
{{prod.kb1}}Answered by AI
{{prod.kb2}}Talk to a human
{{prod.kb3}}New questions to add to the knowledge base
PRODUCT 03 · Data that collects itself and flags problems
Data Workbench
Data scattered across ERP, inventory systems, spreadsheets and platforms flows automatically into one live dashboard. Leaders ask questions directly instead of waiting for reports, and the system alerts the right person before problems hit.
— Live dashboard: sales, inventory, receivables and orders on one screen
— Ask in plain words: “How much did East China collect last week?”
— Key alerts: expiring stock, overdue payments, unusual orders — sent to owners on your schedule
— Numbers are computed by code; AI interprets and advises, it never estimates the figures
Pricing: Quoted by connected data sources and alert rules
Try the expiry alert demo ↓
Business dashboard · liveSample data
Ask{{prod.ask}}
Sales today{{prod.d1}}
Inventory value{{prod.d2}}
Receivables{{prod.d3}}
Expiring within 60 days{{prod.d4}} batches
{{b.d}}
Expiry3 batches expire within 60 days — list sent to the warehouse lead
Receivables2 payments overdue — reps notified
Orders[Product] orders down week over week — awaiting manager review
PRODUCT 04 · An AI chief of staff for the CEO
Executive AI Copilot
Generic AI can make slides and meeting notes. The copilot first captures your experience and judgment, then connects your daily meetings, chats and files, so AI gives advice the way you would.
— Draw out the judgment you “can’t quite explain but always use”, and turn it into rules AI can follow
— Recordings, meetings, chats and files filed automatically, each with its source and expiry
— Before a client meeting, a proposal review or a decision, AI thinks it through your way first
— Correct it once and it remembers — it gets more like you over time
Best for: Owners of law, tax and consulting firms who live on expert judgment, and companies that depend heavily on the owner
Pricing: Monthly retainer, annual contract — quote on request
See how the copilot is built ↓
Copilot · client follow-up workspace● Connected to your files
[Mr. Zhang] said he needs to think about it and has gone quiet since. How should I follow up?
✓ Read client profile
✓ Your past follow-up style
✓ Product and delivery materials
{{prod.brainTyped}}
Data Workbench · expiry alerts · interactive demo
Set how far ahead to be alerted; the system lists which products, how much stock and how much it’s worth. One click sends the list to the owner.
Stock about to expire?Know two months ahead
Alert ahead
Rule: batches expiring before {{inv.untilStr}} ({{inv.rangeLabel}}) · auto-sent every Monday 9:00 · today {{inv.todayStr}}
Expiring batches{{inv.sBatches}}
Products{{inv.sProducts}}
Total quantity{{inv.sQty}}
Cost value{{inv.sCost}}
Value at selling price{{inv.sValue}}
ProductsBatchWarehouseQtyUnit costCost valuePriceExpiry dateLeft
{{r.name}}{{r.batch}}{{r.wh}}{{r.qty}}{{r.cost}}{{r.costAmt}}{{r.price}}{{r.exp}}{{r.left}}
No batches expiring in this range.
AIRecommendations
→{{tip.text}}
How the executive copilot is built,in three steps
STEP {{brain.cur}} / 03Info card #[0217]Auto-generated
ContentAny quote below [X]% of list price must be approved by the owner personally
Source[Date] weekly meeting recording · minute [32]
Applies toQuote approval, sales follow-up
Valid untilRe-confirm after any pricing policy change
copilot/
├─ 01 Me · background, decision habits
├─ 02 Clients · profiles & follow-ups
├─ 03 Business · products, delivery, pricing
├─ 04 Team · roles & collaboration
├─ 05 Industry · policy, competitor moves
└─ 06 Projects · work in progress
index.md · updated as content changes
├─ 01 Me · background, decision habits
├─ 02 Clients · profiles & follow-ups
├─ 03 Business · products, delivery, pricing
├─ 04 Team · roles & collaboration
├─ 05 Industry · policy, competitor moves
└─ 06 Projects · work in progress
index.md · updated as content changes
Confidentiality: Sign an NDA firstKeep materials on your own computer or private environment where possibleBe clear upfront about what it can and can’t do
Use cases by department · 6 departments, 30 use cases
Standard productSupporting serviceCustom on-site build
Beyond the four standard products, here are the use cases we can deliver by department. Green ones use a standard product; dashed ones need an on-site custom build.
{{scene.no}} / 06
{{scene.name}}
{{scene.pain}}
{{scene.std}} use an off-the-shelf product or service
{{scene.fde}} need an on-site custom build
{{c.name}}
{{c.desc}}
{{c.tag}}
Delivery services · FDE on-site delivery
Standard products solve common problems; for processes unique to your company, we send people on-site
Pricing: Business audit — quote on requestOne-department pilot — quote on requestCompany-wide rollout — priced by scopeEach phase signed and accepted separately
Start with use cases like these
✓ High-frequency work that happens every day✓ Close to money: conversion, collections, cost✓ Results you can measure in numbers✓ Can run within one department
Not yet
✕ Low-frequency work done once or twice a month✕ Work that needs several departments to move at once✕ Steps where no error is acceptable — keep a human in charge first
Common first use cases:
Pre-sales customer serviceTop-seller playbook · sales scriptsContent at scaleBusiness dashboardInventory and receivables alertsVoice job reporting on the shop floor · work-order chasingFirst-pass review of contracts and documentsAd creative analysis
Four commitments when we deliver
01Be clear about limitsWhat AI can and can’t do, and what a human must check, is documented before launch.
02Leave the math to codeAmounts, inventory and KPIs that must be exact are computed by code — never estimated by AI.
03Set acceptance metrics firstTargets are based on your historical data and agreed by both sides before work starts — no guesswork.
04It stays with your companyKnowledge bases, workflows, agents and documentation are all handed over; nothing leaves when people do.
FAQ
Do we have to buy all four products?
No. Start with one. All four share one company knowledge base, so products added later reuse it — no need to reorganize your materials.
Can the AI customer service answer everything correctly?
No — no AI gets it 100% right. Our approach: answer what’s in the knowledge base with a citation, hand off to a human when unsure, and agree with you before launch which questions must go to a human.
Does the data workbench replace our ERP?
No. The workbench connects to your existing systems and spreadsheets. Where an old system has no export interface, we assess building one or scheduled exports.
Will our company data leak?
We sign an NDA before starting. For strict data requirements we can deploy on-premises or privately, priced separately. We’ll be clear upfront about what we can and can’t guarantee on security.
How soon can we use it?
A business audit usually takes 2–4 weeks; the first use-case pilot usually 1–3 months, depending on how ready your materials and data are. You get a timeline at the end of the audit.
Pick your most painful use case first.
Magic AssistantOnline · usually replies in seconds
Chat history
No past chats yet. Tap “+” to start a new chat and the current one will be saved here.
{{m.text}}
{{m.text}}
{{m.who}} · Human advisor
{{m.text}}
{{m.text}}
Received, {{m.text}}
Ticket {{m.ticket}}
An advisor will get back to you shortly through the contact you left.
Continue on WeChat with an advisor
On a phone, tap the button to open WeChat; on a computer, scan the code with WeChat.
No WeChat? Just leave a phone number or email here.
People often ask: