# Ask the book a question. Press once on the answer URL: https://www.orla.finance/en/product/assistant Markdown twin of that page. Append `.md` to any Orla page URL to get one. It answers in cards, and nothing moves without your press. #### Cards, one press, a way back It knows what is under the composer: the contact, the account, the row, the bill, the document, the letter, the invoice, the card. A file or a voice note is a question too, and the answer to a figure comes with its chart. - Every write is a card: one card, one press, no model between the press and the ledger, and a path back on each. - The answer is computed from the rows you can see, with your own rights, never from a copy. - What do I pay this week: the signing queue, the calendar's dated obligations and the cash forecast sewn into one order, each row marked fits or move, with the condition named beside the answer: it fits if Acme pays on the 15th. A card postpones one payment and never brings one forward. - How much is really free: three figures for each currency, Free, Committed and Left, where Committed is exactly that week of payments. Wallets and exchanges stand beside the band, never inside it, and nothing here moves money or a date. - Who owes me: one line per client, with one next step and its reason, worked out by Orla rather than by the model. Is this bill a duplicate, how much VAT will the period owe by the day it closes, what got more expensive: each is answered from the rows, and the verdict stays with you. - Attach a statement and say check this against Main: what the statement has and the book does not, what the book has and the statement does not, and how far the totals stand apart. Nothing is written, and which account is a question it asks rather than guesses. - Say connect my Binance, or paste a Google Drive folder link: a Connect button opens that service's panel over the conversation. Keys are typed into the panel, never into the chat, and nothing in a folder is read until you press Approve. - Every page has its own questions under the composer, and a question whose card you could not press is not offered. A section hidden from you takes its tools away from the assistant, and it says so. - The registry is generated from the code: 39 areas of the book, 121 readings, 134 cards it can propose. A connected agent runs forty of the same reports over MCP, with bank details masked unless the connection was approved to see them. For: . #### You see it work, and it asks with buttons - Each step shows as it runs: reading August, comparing by category, building the chart. When a name fits three contacts, it offers the three and Other rather than a question in text. - The finished answer folds the steps into one line, 3 steps, and an opened step says the period and the account it read: Reading cash flow ยท last 3 months. It never repeats a search text or anything typed into it. - The price of reading comes before the reading: when dropped files would take more than half of what is left of the month's AI allowance, Orla asks first, with how many files, about how many tokens, and how many questions are left afterwards. For: . #### A chart when the answer is a series - Figures over time as a line or bars, the cash forecast with a mark on the day the money runs out. A press on a bar asks the next question, and a link beside it opens the rows. For: . #### A list with a box per line Everything the month's close checks, as one list with a counter over it: rows without a category, categories and accounts without a code, payments waiting for a quorum, expenses without a receipt, a day without a rate, statements not filed. Each line has Do it, and the last line closes the month once nothing blocks. - Nothing on the card is stored: the marks are the book as it stands when the card is drawn, so asking again tomorrow draws the list as it is tomorrow. - The last line is offered only once nothing blocks and only where closing is switched on for the space; the press asks to close and a confirmation card follows. The assistant can never reopen a month. For: . #### Forty rows, one press - Where the cards would be many, they come as one list with a box per row and Apply all; each row still goes through the same door one by one, and the answer says what landed and what was refused. For: . #### A message you edit, then send - A reminder to a client, a request for a receipt to a colleague: the assistant offers it as a card, the message is yours to add or change, and the press is yours. It never sends on its own. - Send it mails a late client the same reminder the schedule sends, with the pay link, and refuses when the client asked not to be written to. Ask for receipts reaches the colleague on the channels they allow, with two reminders, on the third and the seventh morning. For: . #### A question that runs on a schedule Every Monday at nine: what is burning, the two-week forecast, who is overdue. A saved question with a schedule and a channel, answered in your own Telegram chat with the bot or by email, with your rights, as prose with one link to the cabinet and no cards. - Any question the assistant can answer becomes a report this way, and stops when you stop it. - Two channels and no others: your own Telegram chat with the bot, never the space's group chat, and your email. Three schedules: every day, a weekday, or a day of the month, never more than once a day. - A routine runs with the rights of whoever set it and is paid from the space's monthly AI allowance; when the allowance is spent it skips the day and says so once. Each plan holds a number of routines, none on the free plan. For: . #### What it remembers, in a list you can read The accountant is Maria, the investor update goes out on the 5th, answer in USD: a note about the space, kept by the owner or an admin, or a note of your own. Nothing is kept until you press Keep. - Only from a card you press or from Settings, never from a phrase in passing: the card shows the words and the scope, and the words are yours to edit before you keep them. A card number, an IBAN, a key or a password is refused before anything is kept. - Up to thirty notes about the space and thirty of your own, listed in Settings under the assistant, where one can be added or dropped by hand. The notes ride the start of every turn as data, with who kept them and when, never as an instruction; ask what it remembers and it lists them, ask it to forget one and the notes come as buttons. Another person's own notes are never shown or used. For: . #### Yesterday's conversation, on any device Every answer is kept as a turn of your own conversation, so continuing it works after a reload and on another device. The clock button lists them, newest first, each named by its first question; pick one and the next question continues it. - The plan sets how long they are kept: seven days on the free plan, thirty on Starter, ninety on Pro, without a limit on the company plans. Forget drops one for good, and Privacy & data counts them and forgets them all. - A conversation is yours alone: the owner of the space does not read a member's, and nobody else's appears in the list. A card restored from an old conversation is shown, not pressed: its figures are old, and a new question brings a fresh card. For: .