AI agent · cross-border hospital & tuition payments
CarryPay is an AI agent that pays your family's hospital and university bills abroad: legally, safely and for less. It deals with the hospital, the bank, the forms and the regulations. You just approve.
Sandbox demo with simulated hospital, university and bank. No real money moves.
Why we built this
“When we went to Bangkok for medical treatment, we carried USD 10,000 in cash because nobody told us Pakistan's rules already allowed up to USD 50,000 to be sent straight to the hospital. When the bill passed USD 10,000, we were stuck.”
The rules weren't the problem. Not knowing them, the paperwork, the fees and the time pressure were. Families in Pakistan, China, Egypt, Nigeria, Bangladesh and many other countries with capital controls face the same thing when they pay for care or education abroad.
It works for days, across the hospital, the bank and your family, until every bill is paid and every refund is back.
It talks to the institution
Emails the hospital or university's verified desk, gets the estimate, the deposit deadline and the bank details.
It knows the rules
Checks your home country's regulations: limits, required forms, direct-to-institution rules. Every rule links to its source.
It picks the best route
Compares bank transfer, card and cash on total cost and arrival time, and explains the trade-off in one sentence.
You approve, it handles the rest
Nothing moves without your fingerprint. Then it tracks the money, fixes returned transfers, checks the final bill and claims refunds.
Measured on the sandbox scenarios, not estimated. Sandbox rates; no real money moved.
PKR 44,060
saved on one hospital case
against paying the same bills by card, plus a THB 14,000 duplicate charge caught and the excess deposit refunded. London tuition: CNY 7,730 saved.
61 actions
handled for the family
over 9 days: 3 emails to the hospital, 11 times the agent picked the case up again. The family's part: approve with a fingerprint.
20 of 20
fake-agent cases stopped
live AI runs of the two fake-admission scenarios (Pakistan and China): nothing paid to the agent, the university checked first, the family told it isn't their fault.
43 of 45
live AI runs passed every check
on the current agent, across all scenarios (details in docs/EVAL.md); 10 milestones of the evidence runs anchored on a public blockchain.
The AI proposes. Deterministic code and your fingerprint decide.
Verified accounts only
Bank details from any email or document are checked against a verified record before money can move.
Untrusted content stays data
Instructions hidden in invoices or look-alike emails are flagged, never followed.
Rules as code
Limits and forms live in versioned rule files, so a policy change is one line, not a new app.
Tamper-evident log
Every action is SHA-256 hash-chained and anchored on a public blockchain at each milestone. Anyone can verify it.
Money needs you
A separate safety layer re-checks every payment the AI proposes. The AI cannot override it.
Emergency stop
One switch pauses everything instantly.
Each country's regulations are a rules file. The agent stays the same.
→
Pakistan → Thailand · medical treatment
SBP: USD 50,000 per case direct to hospital, Appendix V-68 form, card tax 0.5% from July 2026
→
China → United Kingdom · university fees
SAFE: USD 50,000 yearly quota, above it with proof of tuition, no 蚂蚁搬家 splitting
→
Pakistan → United Kingdom · an offer through an agent
No tuition until the university itself confirms the admission; fees only to its own account
→
China → United Kingdom · a 留学中介 promises 保录取
Same check, plus no buying pounds in relatives' names to dodge the quota
Or upload any hospital or university bill from 13 countries: CarryPay prices it under your home country's rules at once, and pays only after the bank has verified the institution.