Everything it does
Every MoneySpark feature is on this page. What is live today is here; what is not is listed separately, so the two never get blurred together.
Three ways to use it
One agent, three conversation modes. Card matching compares the live card database, learning explains financial concepts, advising answers only from the material you give it.
Card match
Describe how you spend and the agent compares reward rates, annual fees and welcome offers across the live card database, then lists the best cards with reasons.
e.g. Groceries and dining — which card pays most?
Start card matchingLearn
Ask about financial concepts without buying anything: built on the IFEC seven-topic framework, with common misconceptions, official page pointers, and every figure dated.
e.g. What happens to existing MPF offset arrangements after abolition?
Start learningAdvisor
Paste in the parts of an insurance proposal, fund KFS or bank statement that you cannot make sense of; the agent sticks to the material you provide, says plainly which fields are missing, and lists what to ask before you sign.
e.g. Help me make sense of a policy proposal
Start with your materialCard matching: from plain words to cards, every step shown
You do not have to learn how to fill in a filter. Say how you actually spend and the agent translates it into a card query, then hands back the results with the reasons attached.
You speak plainly, it queries the database
Cantonese, Chinese or English all work. Say “I order food delivery all the time and travel to Japan every year” and the agent turns that into category reward rates, foreign-currency fees and annual-fee waiver conditions, queries the card database, and ranks what comes back.
Every card comes with a snapshot and a reason
Reward rates, annual fees and welcome offers are shown with the snapshot retrieved at that moment. Card names and every figure come from the lookup — none of them are generated.
Official sources first, never an unsourced number
All card data is read from a live lookup against the card database. When a lookup finds nothing, the agent says it cannot answer right now instead of falling back to a rules engine and inventing a recommendation.
The process is visible while it happens
As the agent works you can see what it searched and how many records it hit. What you see now, what comes back after a refresh, and what the model actually produced are the same thing.
It asks when it is short of detail, and you can reshuffle
When your description is too thin to compare properly, the agent asks a follow-up question instead of guessing. If a batch does not suit you, one tap gives you another set.
Apply from the card itself
Some cards carry an affiliate application link that opens the partner's page, so you are not comparing sites by hand. Not every card has one.
Built-in skills: the playbook behind every answer
Answers are not raw model output. They are driven by a library of built-in skills that a human has reviewed line by line — which is why what it says can be trusted.
Reviewed by a person before it can be used
A model drafts the content, but a person checks every entry — figures especially, against their as-of date — before it is activated. Skills can only call registered commands; they never execute arbitrary code.
It reads a table of contents first
The skill catalogue loads with the conversation and is frozen for its duration; full text is only pulled in once a skill actually matches. That keeps answers explainable and auditable without flooding the context.
It expires, and it says so
Every entry carries an as-of date and a validity window. When it goes stale it is not quietly dropped: the agent still explains the concept but flags that the figures are void and points you to the official page.
Official documents are the source of truth
Official PDFs and licensed-institution pages are the source of truth. What the agent produces is a draft; anything involving money or a signature is yours to decide after reading the official document.
17 built-in skill seeds and growing (as of 2026-10)
The database and the numbers
308 Hong Kong credit cards
Data is maintained and audited by a dedicated team, and the count grows as the database is updated; see the card library for the current number. (As of 2026-10)
Direct application links
Most cards link straight from an answer to the official or affiliate application page, with links audited regularly to make sure they still work. (As of 2026-10)
12 spending categories
Groceries, dining, online shopping, transport, foreign currency and so on — the agent maps what you describe onto the reward rates attached to those categories.
Privacy: the data is yours
Handled under the Hong Kong Personal Data (Privacy) Ordinance, and the core features work without an account.
Visible and deletable
Your profile, memories, conversations and the material you provide can all be reviewed and deleted by you.
Guest data clears itself 7 days after creation
The core features work without signing up. Guest conversations are deleted automatically 7 days after creation, with no renewal; signed-in conversations are kept long term.
We only store the text you paste in
Material you give to Advisor mode is the text you paste in; we never receive or keep the original file.
We do not ask for what we do not need
We do not require personal details that have no bearing on a recommendation or an answer.
Not available yet
The features below are still in development and cannot be used on the live site. They are listed here rather than presented as if they already worked.
Long-term memory and spending profile
Letting the agent remember durable facts about you (“I live on Hong Kong Island”, “I don’t want an annual fee”) and saving your spending profile to your account so it carries across conversations. Running on the test site; the live-site date will be announced separately.
Try it once and see whether the answer was actually looked up.
No sign-up needed; card data always comes from a live lookup, never from memory.