It brings together the data in your business systems and the know-how in your policies, contracts and bids. Staff ask, and every number comes with its source. AI does the repetitive work; people confirm the key decisions.
Staff ask on the web or through a bot in the work chat they already use. The brain works out what kind of question it is, takes the answer from business data or from documents, and says where it came from.
01Business dataQuestions about numbers
How much is still owed on the Chengnan grid project?
How
Filters and calculates exactly over records synced from your business systems.
Answer
Each answer shows record numbers, the data cut-off time and the rule version.
02DocumentsQuestions about rules and practice
How do we get the warranty deposit back?
How
Searches policies, contracts, bids and drawings by meaning.
Answer
Each answer shows the file, page and section; if the documents do not say, it says so.
03Calculation rulesSigned off by finance
Overhead = 5% of contract value (Nov 2026 version, signed off by finance)
How
Each rule states what it calculates, who signed it off, when it applies and its version.
Answer
When a rule changes, old results are kept and new ones start from the effective date.
Business data decides the numbersWhen a document disagrees, both are shown, e.g. “settled ¥4.8M (finance system), contract says ¥5M (contract p. 2)”. AI does not pick one.
Missing data is reported, not guessedIt says which data is missing and who should add it. No zeros, no estimates.
Data sources
Kingdee
DingTalk
WeCom
Feishu
Excel / CSV import
Read-only connections, authorised with your own company accounts.
Examples are illustrative, not customer data.
Real demo
One real question.
Recorded in our demo company: an employee asks the receivables policy assistant, the brain searches the documents they are allowed to see, answers, and lists its sources.
EmployeeA customer is more than 90 days overdue and we cannot collect. How do we write off the bad debt, and who approves it?
Search authorised documentsDone
Procedure and approval (from the Bad-debt write-off procedure (demo) and the Receivables management policy (demo))
Past 90 days overdue, do two things first: refer the receivable to legal for assessment, and set aside a bad-debt provision of 50% of the outstanding balance .
Before applying for a write-off, at least one of these must hold :
legal has given a written opinion that the amount cannot be recovered;
the customer is in bankruptcy liquidation, or litigation or arbitration has ended and enforcement recovered nothing.
Approval, by amount per write-off :
up to ¥500,000: the finance director;
above ¥500,000: the general manager’s office meeting.
After the write-off: the right of recovery is kept, the item is recorded separately in the receivables ledger as “written off, pending recovery”, regional finance reviews it every quarter, and anything recovered reverses the provision .
Two things not to confuse
Waiving overdue interest follows a different rule: up to ¥50,000 by the regional finance manager, above ¥50,000 by the finance director. It does not apply to bad-debt write-offs .
Matching a payment to its invoice on the day it arrives is a different process from a bad-debt write-off .
Not covered by the documents: the write-off application form, the accounting entries, and the time limit for the legal assessment. The current documents do not specify these; please check with the finance department.
Was this answer useful?UsefulNot usefulWrong
Measured in the demo environment (31 Oct 2026): the answer appeared 8.8 s after submitting, the sources at 11.1 s. Translated from the Chinese original; the cited documents are the demo company’s synthetic policies. Playback is shortened.
AI employees
The brain thinks. AI employees do the work.
The same workflow before and after; drag the divider to compare. AI runs the repetitive steps and people confirm the key ones. In the diagnosis we adapt it to how your company works.
We learn which work takes the most people, where the data lives and who signs off, and train your team on what AI can and cannot do.
You receiveA diagnosis: where to start, a suggested timeline and what we need from you
02
30-day free pilot
One piece of work, run in an isolated environment on synthetic or authorised data, to see results, errors and cost.
You receivePilot results to decide on
03
Build your enterprise brain
Connect business data and documents, set up roles and calculation rules, and put AI into the work people already do.
You receiveAn enterprise brain on your own server
04
Handover and adjustments
Operating guide, access settings and monitoring, so your team can run it. When the business changes, we adjust it together.
You receiveA system your team can maintain
Data security
Your data stays with you. What we build is yours to use.
Runs on your own serverOne server per company. Documents, files and keys are kept on your server and our management platform cannot read them; you can also connect your own model service.
Access by role and projectBusiness data and documents are both filtered by each person’s scope; staff cannot see projects they are not allowed to.
Original records are read-onlySynced records and uploaded originals are kept read-only; corrections are saved as new versions, so every number traces back to its record.
People confirm key actionsConnections to other systems are read-only by default; anything that writes or approves needs a person’s confirmation and is logged.
No estimates, no forced answersWhen data is missing it says what and who should add it; when the documents do not say, it says so.
Yours to keep usingThe business logic, configuration, rules and knowledge-base structure built for you stay in use by your company.
Industry templates
Companies in the same industry start from a template.
Each industry keeps the business objects and rules it has in common, so a new company only configures its own record mappings, names and parameters instead of starting from zero. Workflow examples are available for these industries:
The partners beside us, and the technology we build on.
Products we build and run ourselves, the organisations we work with, the infrastructure behind our delivery, and the models and platforms we actually use and integrate.
Third-party names and logos are trademarks or registered trademarks of their respective owners. "Close partners" means organisations we have an actual business relationship with. Listing under "Infrastructure & domains" or "Models & platforms" means only that we use their products or services; it does not imply endorsement, sponsorship or partnership. Cloudflare and the Cloudflare logo are trademarks and/or registered trademarks of Cloudflare, Inc. in the United States and other jurisdictions.
Free diagnosis
Start with a free diagnosis.
Tell us which work takes the most people. We learn how your business and data run, train your team on AI and suggest how to build your enterprise brain, free of charge. When you are ready, start with a 30-day free pilot.
