Skip to main content
The Accounting Intelligence composer accepts spoken prompts as well as typed ones. You record from the microphone in the composer, Accounting Intelligence transcribes what you said into the draft, and you review that text before sending it. The microphone is available in the web app and the desktop app, on the same composer you use in the Agents panel and on the full Accounting Intelligence page. Dictation does not send the prompt. It only fills the draft, so you can fix a name, add a date, or attach a file before Accounting Intelligence runs.

​
Before you start

Voice input uses the same Accounting Intelligence composer you already type into, so check these prerequisites:
  • Access: Accounting Intelligence must be available on your organization and role. If you cannot open the Agents panel, you cannot dictate into it either. See user roles and permissions.
  • Microphone: Connect a working microphone to the computer you are using. Accounting Intelligence records from the system default input.
  • Permission: The browser or desktop app must be allowed to use that microphone. The first time you record, it asks. If you dismissed that prompt, see Allow microphone access.
  • An open conversation: Open the Agents panel with Option+Cmd+] (Mac) or Ctrl+Alt+] (Windows) by default, or go to Accounting Intelligence from the top bar. Both shortcuts can be rebound; see Keyboard Shortcuts.
There is no keyboard shortcut for starting or stopping a recording. You start and stop from the microphone control in the composer. Set the entity selector and the accounting-basis selector before you record a question that returns a balance. Those selectors scope the whole conversation, including a prompt you dictate. See How Accounting Intelligence Works.

​
Dictate a prompt

Recording starts from the microphone in the composer, stops when you choose Stop recording, and then Accounting Intelligence transcribes the clip into the draft. Follow these steps each time you want to speak a prompt.
  1. Open Accounting Intelligence and place the cursor in the composer where the transcript should land.
  2. Choose the Voice input microphone at the bottom right of the composer, next to send.
  3. Allow microphone access if the browser or the desktop app asks.
  4. Speak the prompt. While Accounting Intelligence is recording, the microphone is replaced by a timer and the Cancel recording and Stop recording controls, and the composer stays editable so you can keep typing.
  5. Choose Stop recording to stop. Accounting Intelligence then transcribes the recording. A spinner replaces the microphone until the text lands in the draft.
  6. Read the transcript, edit it if needed, and send it with the arrow control or the keyboard.
After you send it, the prompt appears in the conversation and Accounting Intelligence answers it like a typed prompt. A recording can last just under two minutes. The timer changes color as you approach that limit, and Accounting Intelligence stops the recording automatically before it hits two minutes, then transcribes what you already said. Start a second recording if the prompt is longer. To discard a recording instead of transcribing it, choose Cancel recording next to the timer, or press Esc while Accounting Intelligence is recording. Navigating away also discards the clip. Cancel does not upload audio. Only stopping a recording, or hitting the automatic time limit, sends the clip for transcription. You can start a recording while Accounting Intelligence is still answering a previous prompt. Dictation edits only the draft. Send stays gated the same way it does when you type, so a response in progress is not interrupted. The microphone is hidden when the composer itself is disabled.
Accounting Intelligence detects the language you speak and transcribes with accounting vocabulary in mind, so accounting terms such as accruals or deferred revenue are more likely to come out correctly. Still read the draft, especially vendor and account names. A misheard vendor or period is easier to correct in the composer than after you send.

​
Review the transcript before you send

The transcript becomes the text of your draft. Stopping a recording sends the audio for transcription, but the transcribed prompt does not become a conversation message until you send it. Treat the inserted text the same way you treat a sentence you typed. Accounting Intelligence inserts the transcript at the caret, with spacing so it does not run into the words on either side, and moves the caret to the end of the inserted text. If you already had a half-written prompt, the spoken sentence lands where you left the cursor. If you never focused the composer, the transcript is appended. Edit the draft after transcription the way you would any other prompt. Fix a misheard account, add a date range, or invoke a skill by typing / as the first character. Attachments, the entity selector, and the accounting-basis selector are unchanged by dictation. Send the prompt only when the wording is the question you meant to ask. A failed transcription does not clear the draft you already had. The clip stays available for Retry until you dismiss the error or start a new recording, so you do not have to speak the prompt again to recover from a network or transcription failure. For how a sent prompt is grounded in your ledger, and what Accounting Intelligence may change after you send, see How Accounting Intelligence Works. For how the audio itself is handled, see What happens to audio I dictate into Accounting Intelligence.

​
Allow microphone access

Voice input needs permission to use the microphone, and the composer cannot record until that permission is granted. The first prompt is a browser or operating-system dialog. If you blocked it, the composer shows Microphone access is blocked on the microphone control and leaves the control available so you can try again after you allow access. In the web app, the popover walks through the browser setting:
  1. Choose the lock or settings icon in the address bar.
  2. Set Microphone to Allow.
  3. Choose the microphone in the composer again.
In the desktop app, the block is usually at the operating-system level rather than in a browser site setting. The popover tells you to allow the app to use the microphone in System Settings on macOS, or to turn on microphone access for desktop apps on Windows. Choose Open Settings to open that page, allow the app, then return and choose the microphone again. The desktop app also asks at the operating-system level the first time you record, and only then. It does not prompt for the microphone when the app starts. If you denied that first prompt, Open Settings is the path back. Camera access is not part of voice input. Accounting Intelligence does not request a camera to transcribe a prompt.

​
Troubleshooting

Most recording failures are permission or device problems, and a failed transcription can be retried without speaking again. The table below maps what you see to what to do.
Last modified on November 5, 2026