Ovirno. Try it on your data
01

Voice

Run your business from a voice note.

Some owners live in dashboards. Others want to be anywhere but. Send a voice note in Telegram and get an answer computed from your own orders — no app to learn, no screen to open.

  • up to 5 minutes per note
  • transcribed in the EU
  • recognition set to Russian
  • 300 minutes of speech a month
  • 200 questions a month
02

What actually happens

Four steps, and the bot narrates every one of them out loud.

  1. you pressA voice note in Telegram, in your own words. The bot answers “listening…” before it does anything else.
  2. secondsThe recording goes to the transcriber and comes back as text. The bot edits its own message to show you exactly what it heard — so a misheard word is caught by you, not buried in the answer.
  3. under a minute“Computing from your data — usually under a minute…” is what it says next, because that is what it takes. The answer is computed from your own orders and your own metrics, not from a cached report.
  4. your callThe answer is text: read it and decide. If you want it as work, press “new task” and say it — a second later it is on your own list, and the bot repeats the task back so you can see what it wrote down. Anything that lands on somebody else's list goes through the web, with a name, a date and a button; the bot never does that from a voice note.

Numbers like “recognition in under a second” are not printed here because there is nothing in the code that measures them. What is in the code is the sentence the bot says to you, and it says “usually under a minute”.

03

What it does today

Working now

  • It answers questions about your numbersSales, orders, customers, tasks — computed live from your connected store and sheets while you wait, not read out of a canned report. The answer says how it was counted.
  • It searches what you have filedDocuments, connected mailboxes and chat history are searched for supporting passages and quoted back. Text from those materials is handed to the model as data, inside a frame that says so — a forwarded letter must not be able to give orders.
  • It turns a recorded call into drafted tasksPaste the transcript and it extracts who agreed to do what, with the quote each agreement came from. Nothing is created until a person confirms.
  • It takes a task — for you, by voiceHands full, phone in one hand: press “new task”, say it out loud, and it is on your own list a second later, with the origin recorded. Only your own: putting work on somebody else's list goes through the web, with a name, a date and a button. A task you gave yourself by mistake is a shrug; one you gave a colleague by mistake is a phone call.

What it honestly doesn't

  • It doesn't write to your customers because it felt like it.Mail that leaves the system on its own — booking confirmations, the reminder the day before, the review request after delivery — goes only after you switch that channel on, and it is off until you do. Then it is the switch that decides, not a button on each letter. Nothing is ever sent to a customer from your voice note.
  • It doesn't reply with a synthesized voice.Answers come back as text you can skim. Nobody wants to listen to a paragraph of numbers.
  • It doesn't take dictation longer than five minutes.Past that it is a meeting recording, not a command — upload it in the meetings section instead, where the whole thing is turned into tasks.
  • It doesn't invent numbers.A day with no sales is reported as a day with no sales; missing data is reported as missing.
  • It doesn't work without the recognition key.In a private chat it says so and asks you to type instead. In a group it stays quiet — a bot that answers every voice note in a team chat with an apology is worse than a bot that waits — so a note dropped into a group when recognition is off, or one longer than five minutes, goes unanswered. That is the one place where silence is the answer, and it is written here rather than left to be noticed.
04

Where your voice goes

Named provider, named country, and nothing kept afterwards.

  1. Voice noteTelegram
  2. Server memoryGermany
  3. TranscriptionMistral · Sweden
  4. Answerfrom your own numbers

the recording is never stored — there is nothing to delete

  1. Your data lives on our server in Germany. The company behind it is Kodawari OÜ, Tallinn.
  2. Voice notes are transcribed by Mistral — a French company, processing in Sweden with Ireland as the fallback. The provider sits behind an interface in our code, so changing vendor is a line in the environment rather than an excavation.
  3. The recording is not stored. It is fetched from Telegram into memory, sent to the transcriber and gone. There is no “deleted after thirty days” here, because there is nothing kept to delete. What is written down is the number of seconds — that is what the spending cap counts.
  4. There are two caps, and both are in code rather than in a policy. Recognition: three hundred minutes a month per company — reached, and the bot says so and asks you to type. Questions to the analyst: two hundred a month per company, counted from the run log, and a voice question spends one of them like any other. Reached, and the analyst says the month's questions are used up while the dashboard, the observations and every section carry on as usual. Both are fuses on the bill, and both are printed here rather than found on the day you reach them.
  5. The recognition language is set explicitly to Russian, and the analyst answers in Russian. Left to guess, it decides a sentence with a brand name in it is Lithuanian and the whole transcript goes in the bin. The workspace itself reads in English, Russian, Ukrainian and German — the machine's own sentences have not caught up yet.
  6. What their contract says about training on your audio is their document, not our repository — so it is not promised here. Ask, and we send you the provider's terms rather than a claim of our own.

One step

Hear it answer your numbers.

Bring one export from your store or a sales sheet. We connect it, and you ask the first question out loud the same day.

Show me my numbers Or write on Telegram

If speech recognition is not switched on for your installation yet, you will be told that in the first minute rather than the second week.

Try it on your data