Voice input for transactions

You can dictate an expense or income by voice — the app recognizes speech, automatically fills in the amount, type, date, and account, and you just need to review and confirm. Available in the mobile app and on the web (microphone button on the add form).

How to use

1. Tap the microphone and dictate the transaction in plain language:
   • “spent 500 rubles on groceries yesterday”
   • “payé 30 euros au supermarché”
   • “20 Euro für Kaffee”
2. Recording stops automatically (up to 20 seconds — enough for one transaction).
3. The app will show a draft with the recognized fields. Nothing is saved yet.
4. Make any necessary corrections (amount, account, category, date) and tap “Confirm” — only then will the actual transaction be created.

What is recognized

Amount, type (expense/income/transfer), currency, date, vendor/description, and suggestions for account and category.

• Date. The app understands relative expressions (“today,” “yesterday,” “the day before yesterday,” “tomorrow”) and specific dates (“May 14,” “06/24/2026”). Relative dates are calculated based on today’s date in your time zone, so “yesterday” always means yesterday regardless of what the engine recognized. If you don’t state the date, it defaults to today.
• Currency. If you don’t specify the currency, your default currency (from your profile settings) is used.
• Account and category. The app suggests these from your existing accounts and tags (“cash” → “Cash” account, “groceries” → “Groceries” tag). If there’s no match, the field remains empty — the app doesn’t create a new account on its own; you choose it in the draft. Any field the app is unsure about (e.g., no amount, unclear account) is marked as “needs review” — so always check the draft before confirming.

Languages

Russian, English, German, Spanish, and French are supported — the language is detected automatically, no switching needed. You can mix them (e.g., Russian speech with an amount in euros).

Limitations and privacy

• Length — up to 20 seconds. Longer recordings are rejected: one short transaction at a time, not a list dictation.
• Plan limit. Each recognition consumes one unit of your plan’s monthly voice input quota (one tap = one unit, even if the app tried several recognition engines under the hood).
• Recognition is an external service. Audio is sent to a third-party speech engine solely for recognition; the draft is created from the text. A real transaction is not created without your explicit confirmation.

If it didn’t work

• “Doesn’t look like a transaction.” The recording lacked an amount or a clear operation — speak more clearly, for example, “spent 500 rubles on …”
• “Could not recognize” / service unavailable. A temporary issue on the recognition engine’s side — try again in a minute. The quota is not charged in this case.
• “Limit reached.” The monthly voice input quota is exhausted — enter the transaction manually or upgrade your plan.
• “AI temporarily unavailable.” On the free plan, AI features also have a daily service limit (protection against abuse). If this limit is reached for the day, recognition/checks are temporarily unavailable — enter the transaction manually, and the limit will reset tomorrow. Paid plans do not have this restriction. The quota is not charged in this case.

See also: ai-transaction-drafts.md (general AI draft flow), voice-transcription-pipeline.md (technical description of the recognition engine).

---