Open Chat and ask your books a question in plain language — "who owes me money?", "what did I spend on travel last month?". The assistant reads your live data, answers with numbers and charts, and never writes anything without showing you a confirmation card first.
What can I ask the assistant?
Every ability below is limited to your own workspace's data, and reads it under your own login — so an assistant or an accountant asking sees exactly what they would see on screen:
- Find a client and pull up their details, projects and recent payments.
- List payments or expenses filtered by client, category, mode or date range.
- Summarise a period — income, expenses and the picture for any month or range.
- Work out who owes you — outstanding balances across clients.
- Draw a chart — answers can include inline charts, not just tables of numbers.
- Search your site visits — plain-language search that also matches Hindi, Tamil, Telugu and Kannada words.
- Show your sales funnel — deals by pipeline stage.
- Summarise partner commissions — what's accrued and payable.
- Simulate a scenario — "what if I raise prices 10% and volume drops 5%?" projected from your recent actuals.
- Propose an entry — draft a payment, expense or visit for you to confirm.
- Look up a record or a document — what you hold for a customer under your own record types, and what is about to run out.
- Find the right person at a company — the roster at an account, filed by role rather than by job title.
- Review calls and website chats — what was dialled or missed, and the conversations the chat widget and WhatsApp responder held.
- Read back saved research — facts somebody saved about a customer, each with the page it came from.
- Check your outreach — WhatsApp template approvals, what WhatsApp is costing this month, email opens and clicks, and which ads produced leads.
- Review your Tally syncs — what each run covered, what it would post, and what needed a person.
Three things it deliberately cannot reach: your mailbox, your saved logins, and the exact copy a customer would receive from a template. The first two need credentials only the server may hold, and the third would show you a preview that could quietly disagree with what actually goes out.
While it works, you'll see small pills showing which ability it's using — nothing happens invisibly.
Does the AI ever change my books on its own?
No — this is the assistant's hard rule. Anything that would write (add a payment, an expense, a visit) is only ever proposed: the assistant shows a confirmation card with every field filled in, and nothing is recorded until you tap confirm. Decline the card and nothing happens. Reads are just reads.
How do I use voice input?
Tap the microphone in the chat box and speak — your speech is transcribed and dropped into the input for you to review and send. Handy on the shop floor or in the car park between site visits.
How do conversations work?
Chats are saved as threads — come back to one and continue where you left off. Responses stream in live as they're generated. The assistant carries your last 30 messages of context in a thread, so long conversations stay focused on the recent discussion.
What are the limits?
Fair-use caps keep the included AI free for everyone: up to 50 messages per day per person, up to 8 data-lookups per answer, and answers capped at a sensible length. If you hit the daily cap, it resets the next day — or bring your own AI key to run on your own provider account.
Can it read a receipt photo?
That's the assistant's sibling, Scan-to-add: photograph a receipt or bill and Finocket extracts the amount, date, GST and category into a draft expense or invoice — also confirm-before-save. See the Scan-to-add pointer on the + menu.
Related: Bring your own AI keys, Usage & activity.
What Mysty will not say about money or tax
Every sentence Mysty writes about money or tax law is checked against the figures it was actually handed in that answer. A sentence quoting anything else is dropped before it reaches you — you will never see a number Mysty worked out on its own.
- When the tax working refuses, Mysty refuses with it. It tells you plainly that it stopped and why, and it does not fill the silence with a figure — “₹0” included, because nothing logged is not the same as nothing owed. That now holds on the written summaries on your tax screens too, not only in chat: a zero there supports no sentence at all, so a nil liability cannot be stated. Real amounts are unaffected — ₹40,000 contains zeros; only a figure that is nothing but zeros is refused.
- It quotes a statutory limit only from the reviewed rates for a year Finocket has read, and only for the year you asked about.
- It will not tell you which tax regime to elect or what to invest in — and that line holds in Hindi, Tamil, Telugu and Kannada, not only English. It holds wherever a tax figure is explained too, not just in chat: the same refusal now applies to the written summaries on your tax screens.
- If nothing it wrote could be stood behind, it says so in one line and points you at the screen with the real figures, rather than showing you a confident paragraph.
The same rule applies to what Mysty drafts on your behalf
Checking what Mysty says to you is only half of it. When Mysty proposes an action — an email to a customer, a WhatsApp reply, a payment reminder, an invoice note, a journal narration — that text is written by Mysty too, and it leaves your business.
So every figure in a drafted message must come from one of two places: something Mysty actually looked up while answering you, or a number already on the confirm card in front of you (an amount, a quantity, a rate). A figure from neither means no confirm card appears at all — the draft is refused whole and Mysty tells you why.
It is refused rather than quietly corrected on purpose. Deleting a sentence from an email you are about to approve would leave you tapping Confirm on a message you never read, and possibly one that now means something different.