Your cards. Your destination. One verified money plan. Tell us where you are going and what you already carry, and we will assign each card a role, size your cash, and give you three things to do before you fly.
Most travel-money advice is written for a generic traveler. This tool is not. It starts from the payment methods you already own and asks a narrower question: for this destination, this trip length, and this spending pattern, does your current wallet work, and what is the smallest change that would make it work better?
It assigns a role to each payment method you describe. One card should carry ordinary purchases. One card should handle cash withdrawals, if you need cash at all. One card should sit untouched as a backup for the day something gets frozen, swallowed, or stolen. Most travel-money problems come from a wallet where two of those roles are missing or accidentally assigned to the same card.
Every number in your plan is produced by our own code before any AI is involved. The engine reuses the same fee model as the True Cost of Travel Calculator: a typical home-bank international ATM fee, a per-destination ATM operator surcharge, a conversion markup, and the dynamic currency conversion markup that appears when a terminal offers to bill you in your home currency.
From your expected trip spending we derive a cash range for the destination, convert that into a likely number of withdrawals, and price each withdrawal against what your ATM card actually does. Where a fee is unknown we show a range bounded by "no fee" at one end and our published planning heuristic at the other, rather than inventing a rate. The result is a modeled avoidable fee exposure with every assumption listed underneath it — a planning range, not a quote, and not a saving you are promised.
Exactly one language-model call happens per plan, and only after all the arithmetic is finished. Its job is judgement and explanation: is this wallet already sufficient, which method should play which role, what is the smallest useful change, and what deserves your attention before you fly.
It is not allowed to calculate money, invent issuer terms, add card products that you did not tell us about, predict credit approval, or promise you a saving. It never sees our paid product catalog, so it cannot recommend a purchase. Every response is checked by a deterministic verifier that rejects fabricated figures, hallucinated card names, absolute claims, and any output that quietly drops an uncertainty we already recorded. If that check fails, the AI layer is discarded and you get the deterministic plan.
There is no account and no saved profile. We never ask for a card number, account number, PIN, balance, credit limit, routing number, password, passport, or tax identifier, and the form refuses anything that looks like one.
To be precise about where your answers go, because “nothing is stored” would be too broad a claim. Travel Card Index does not persist them: this service has no database, and your request body is never written to our application logs — we record only a request ID, the destination, how many payment methods you described, timing, and whether the AI layer was applied. They are processed in transit: your answers go to our backend to be analysed, and a compact summary of the resulting facts is sent to our configured AI provider (Groq, running openai/gpt-oss-20b) for the single judgement step. We do not control that provider’s retention, and we do not make claims on its behalf — see the provider’s own terms for how it processes API content. That is exactly why the form asks you never to enter card numbers, account numbers, PINs, balances, or passport details. Read the full approach and our editorial policy.
Three of the four fee layers are outside our knowledge and sometimes outside yours. ATM operator surcharges vary by machine and change without notice. Conversion markups depend on your specific agreement. The DCC markup depends on whichever processor the merchant happens to use. We publish a per-destination planning model for these, clearly labelled as an editorial model rather than an issuer quote, and we tell you when a fact behind your plan is old enough to re-check.
Card terms in our catalog come from issuer and provider documentation only — never from affiliate blogs or aggregator pages. Each one carries the source URL and the date it was last checked, and your result shows both. A fact checked within 90 days is treated as fresh, one between 90 and 180 days is due for review, and anything older is flagged as stale with an explicit instruction to verify the issuer's current terms before relying on it. Our running record of every named-product financial claim on this site is kept in the source register we maintain alongside the content.
Affiliate status cannot influence what this tool tells you, and that is enforced structurally rather than as a promise. The scoring code has no access to the affiliate flag at all, and an automated test fails our build if flipping that flag changes any recommendation, any ordering, or any verdict. Affiliate metadata may change where a link points and what disclosure appears next to it. It may never change what we say is right for you.
No. The calculator prices one ATM withdrawal in isolation. Travel Wallet Lab takes the cards you already carry and assigns each one a role for a specific trip: which card buys, which card withdraws, which card is the backup, and how much cash the destination actually needs.
We do not store them. There are no accounts and no saved profiles, this service has no database, and the request body is never written to our application logs — we log only metadata: a request ID, the destination, how many payment methods were described, and timing. Your answers are still processed in transit: they reach our backend, and a compact summary of the resulting facts is sent to our configured AI provider for the single judgement step. We do not control that provider's retention and make no claims on its behalf, which is why the form asks you never to enter card numbers, account numbers, PINs, balances, or passport details.
No. Every figure is a USD-equivalent planning estimate built from your inputs, our destination payment research, and the same editorial fee model the calculator uses. Where an issuer term is verified against official documentation we say so and link the source. Where it is unknown, we say unknown rather than guessing.
It performs judgement only, and only once per plan. It never calculates money, never invents card terms, and never chooses which paid guide you are shown. All arithmetic happens in our code before the model is called, and every model response is checked by a deterministic verifier before you see it. If that check fails, or the service is busy, you get the deterministic plan instead.
No, and that is enforced in code rather than promised. The recommendation engine cannot read affiliate status at all, and a regression test fails the build if toggling that flag changes any recommendation, ordering, or verdict.
The free beta service sleeps when idle, so the first request after a quiet period has to wake it up. Nothing is wrong; the plan will arrive.