Just ask where your money is - and get proof.
A read-only AI over your ledger. It asks tools, then explains - the numbers never pass through the model, so it cannot make them up. Ask in English or Roman Nepali; get the rows behind the answer.
The AI is deliberately thin and impossible to be wrong about money. The iron rule is architectural, not aspirational: the ledger computes the numbers; the model asks read-only tools, then explains. There is no code path from model output to your money.
That means hallucination is structurally impossible for the figures - only the words pass through the model. Ask 'khalti bata last month kati aayo?' and you get an exact number, the count of matched payments behind it, and a link to the rows.
What you get
- Read-only tools over the ledger - close, payments, settlements, fees, exceptions
- Every figure tagged with its evidence
- English, Nepali and Roman Nepali understood natively
- Refuses rather than estimates money it can't prove
- One gated action - draft a refund for a human to confirm
- No cross-organization data, ever
Ask your money, in detail.
The iron rule
The iron rule is architectural, not aspirational: the ledger computes every number, and the model only asks read-only tools and then explains what came back. There is no code path from model output to a figure, so hallucination is structurally impossible for money - only the words pass through the model. Ask something the ledger cannot answer and it refuses, rather than inventing a number to be helpful.
Read-only across the whole ledger
The AI can reach every part of your reconciled ledger - the daily close, payments, settlements, fees, refunds and the exception queue - but only to read. Its toolbelt is queries, not writes, so a question can never change a balance. Ask 'khalti bata last month kati aayo?' and it gathers the matched payments and returns the exact total.
Every figure carries its evidence
An answer is never just a number. Each figure comes tagged with where it came from: the count of payments behind it, the settlements it spans, and a link to the exact rows. You can open any claim and see the ledger lines that prove it, so the AI stays something you check rather than something you trust on faith.
Ask in the language you think in
The chat understands English, Nepali and Roman Nepali natively, so 'hijo ko esewa settlement kati aayo?' works exactly as typed. Answers can come back in the warm Roman Nepali register you use on Viber, while anything you export for your accountant stays in formal English. Language is a simple per-user setting, not a mode you have to remember to switch.
One gated action, no cross-org data
The model has exactly one action beyond reading: it can draft a refund, which then waits for an authenticated human to confirm behind the refund role and its cap. It can prepare the work, never execute it. And it only ever sees your organization's ledger - there is no query, prompt or path that reaches another business's data.
FAQ
Ask your money,
explained
Common questions about ask your money. Want it run on your own data first? The concierge reconciliation is free.
Not the figures. The ledger computes them and the model only restates tool outputs - so numbers never pass through the model. It refuses rather than estimates.
Get your free reconciliation.
Send us last month's eSewa report, Khalti export, bank statement and orders sheet. Within 48 hours we send back a one-page close - every payment matched, every fee computed, every settlement decomposed.
Files only · private upload link · nothing to install