AI assistant
An AI in your books you can preview, cap and undo
Ask 'who owes me money?' in plain language — and when it offers to act, see the money before you confirm, cap how far it may go, and reverse what it posted.
What it solves
The answer you need — who owes you the most, what your GST is this month, how your funnel looks — is buried in reports you'd have to open and read. But the reason most owners won't let an AI near their accounts isn't that it can't find the number; it's that nothing tells them what it's about to do, and nothing undoes it afterwards. Finocket answers both: the assistant reads your live data and tells you, and every write it proposes arrives with its money impact on the confirm card, a ceiling you set, an append-only record and — for a posted journal — a one-tap reversal.
How we're different
The controls are the product, not a disclaimer. You see the money before it moves; you set the limit in rupees and in how many people one action may reach, never as a confidence percentage; every action lands in a log Postgres itself chains with a hash so tampering is detectable; and a sentence stating a figure the model wasn't handed is deleted rather than softened. Prefer your own provider? Bring your own Anthropic (Claude), OpenAI or Google AI key, stored encrypted, and the assistant uses it instead of the included AI.
Inside AI assistant
Answers from your own data
See the money before you confirm
AI assistant
See the money before you confirm
An autonomy dial in rupees and inboxes
AI assistant
An autonomy dial in rupees and inboxes
Undo after it posted
AI assistant
Undo after it posted
A log the database makes tamper-evident
AI assistant
A log the database makes tamper-evident
AI assistant
See the money before you confirm
AI assistant
An autonomy dial in rupees and inboxes
AI assistant
Undo after it posted
AI assistant
A log the database makes tamper-evident
More in AI assistant
One door for every write
All 20 things it can do — record a payment, draft an invoice, post a journal, import a statement, raise a PO, record a vendor bill, send a reminder, launch a campaign, approve a GST filing — go through one resolve step that validates, executes through the poster that already existed for that record type, and appends the audit row. The AI path gets exactly the validation, permissions and ledger posting the manual path gets, and no more.
A figure it can't cite is deleted
Each finished sentence about your numbers is checked against the figures it was handed; one it cannot cite is dropped, not hedged — mid-answer in chat, and identically on report summaries, the insight digest, outreach drafts, WhatsApp replies and GST-notice drafts. It governs what the assistant SAYS to you; free text inside a proposed action is still yours to read before you confirm.
Included, metered — or bring your own key
The assistant is included on every plan, free but metered: 50 messages a day across the workspace, with 100 document scans and 60 voice notes a month. Add your own Anthropic (Claude), OpenAI or Google AI key at the company or personal level — stored encrypted, never shown again — and those caps come off.
Works with the rest of Finocket
Part of AI you can preview, cap and undo. See how every module connects.
Questions
What can I ask it?
Plain-language questions about your books and pipeline — 'who owes me the most?', 'what's my GST this month?', 'how's my funnel?' — and it answers from your live data with links, confirming before it writes anything.
What stops it doing something expensive and wrong?
You see the money impact before you confirm — how much moves, out of which bucket, how many people it reaches, whether it's undoable. You set a ceiling in rupees and in recipients, and no setting removes the human confirm. Everything it does lands in an append-only log the database chains with a server-computed hash, so tampering is detectable. And a journal it posted can be reversed afterwards as a linked compensating entry rather than a deletion.
Can it make up a number?
It's built so it can't state one. Every finished sentence about your figures is checked against the numbers it was actually handed, and one it can't cite is dropped rather than softened — in chat and identically on report summaries, the digest and the drafts. The limit worth knowing: that guard covers what it SAYS to you, so the free text inside an action it proposes is still yours to read before you confirm.
Can I use my own AI provider?
Yes — add your own Anthropic (Claude), OpenAI or Google AI key at the company level for everyone or a personal key just for you. It's stored encrypted. Set no key and Finocket's included AI is used: free, but capped at 50 messages a day, 100 scans and 60 voice notes a month. Your own key removes the caps and unlocks the larger model.
Learn how it works
Bring your own AI keys
Run Finocket's AI features on your own Anthropic, OpenAI or Google account — personal key first, then company key, then Finocket's included AI.
Notifications, and who gets them
The bell tells you what needs you, and now it reaches your team rather than only the owner. Turn categories off, set quiet hours in your own timezone — security alerts always come through.
3 min read
Closing your Finocket account
Close your account yourself from Settings → Account & privacy. Nothing is erased for 30 days — a grace window you can cancel in; export your CSV first, because after that there is no copy to ask for.
3 min read
Ready to ditch the spreadsheet?
Free for solo users. No credit card. Your books stay yours.