Skip to main content
Xuula

Where your words go

This page describes what actually happens, step by step. It is written to be checked against the product rather than to reassure. The Privacy Policy is the formal version.

1On your PC

  • Detecting the hotkey and capturing microphone audio — and, during a meeting recording, Windows system audio.
  • Showing the recording indicator and the live words as they settle.
  • Inserting the finished text into whichever field has focus.
  • Storing your dictation history. In this release, history is local only and is not uploaded.
  • Storing meeting recordings, transcripts, translations, notes, speaker names, word timestamps and AI insights — all of it in local files, organised into Projects. None of it is uploaded to Xuula servers.
  • Storing your personal dictionary and per-project vocabulary locally so they can be applied to recognition.

2Sent to cloud services

  • Audio, while a dictation or a meeting recording you started is active — for meetings, the microphone and, when enabled, system audio

    → Speech recognition provider

    To convert speech to text, including language switches, speaker separation in meetings, and optional translation.

  • The transcript text, plus your selected style and relevant dictionary terms

    → AI provider

    To remove filler words, fix grammar and apply punctuation. Skipped entirely when cleanup is off.

  • The speaker-labelled meeting transcript text — never the recording

    → AI provider

    Only when you request an AI meeting action: a summary, key points, to-dos, quotes, or an assistant answer.

  • A word count, a device identifier and a timestamp

    → Xuula servers

    To enforce the free plan’s weekly allowance and to keep an auditable usage record. The words themselves are not sent.

  • Email address and subscription state

    → Xuula servers and the payment processor

    To run your account and take payment. Card details go directly to the payment processor.

What is not sent to our servers

  • Your audio — dictation or meetings. It goes to the speech provider, not to us, and we do not store it.
  • Your transcripts, meeting recordings, notes and Projects. Usage reporting sends a count, not the content.
  • The contents of the application you were typing into.
  • Your full IP address as a stored record. We keep at most a country code.

Providers involved

Each provider has its own retention and model-training terms. Those terms are stated in the Privacy Policy only where they have been confirmed against a signed agreement — anywhere they have not been, the page says so rather than guessing.

  • Soniox

    Terms not verified

    Speech-to-text transcription, including mixed-language recognition, meeting transcription with speaker separation, and optional translation.

    • · Microphone audio captured while dictation is active
    • · During a meeting recording: microphone audio and, when enabled, Windows system audio, mixed into one stream
    • · Language hints and, when configured, personal-dictionary terms and per-project vocabulary used as recognition context

    How long Soniox keeps audio, and whether it may use it to improve its own models, is governed by the Soniox terms and the account agreement in force — not by this policy. We have not verified those terms against a signed agreement, so nothing here should be read as a promise that your audio is excluded from retention or from training. Soniox’s own privacy terms are what it commits to.

  • OpenAI

    Terms not verified

    Optional AI processing: dictation cleanup (grammar, filler removal, punctuation, writing style), final meeting-transcript polish, and on-demand meeting insights (summary, key points, to-dos, speaker quotes, assistant answers).

    • · The raw transcript text of the current dictation
    • · The selected writing style and personal-dictionary terms relevant to that dictation
    • · When an AI meeting action is used: the speaker-labelled meeting transcript text (never the audio)

    How long OpenAI keeps the text sent to it, and whether it may use it for training, is governed by the OpenAI API terms and the account agreement in force. We have not verified those terms against a signed agreement, and zero-data-retention is not confirmed for this account — so nothing here should be read as a promise that your text is excluded from retention or from training. The text now reaches OpenAI through Xuula servers rather than directly from the Windows application — nobody supplies a key of their own any more — and each request asks OpenAI not to retain it. Xuula does not store the text: it is held for the length of one request and dropped. Audio is never sent to OpenAI, and turning AI cleanup off stops text being sent at all.

  • Stripe

    Payment processing, subscription management, invoicing and tax calculation.

    • · Email address and account identifier
    • · Billing details and payment method entered directly into Stripe
    • · Subscription and invoice records

    Card numbers never reach Xuula servers; they are entered into Stripe-hosted Checkout. Stripe acts as an independent controller for fraud prevention and regulatory purposes, and keeps payment records for as long as its own obligations require. Payments are switched off on this deployment, so no customer data has been sent to Stripe yet.

  • Railway

    Terms not verified

    Hosting the website and running the application server. The production service runs in Railway’s us-east4 region, in the United States.

    • · Every request to the site, including IP address and request metadata in the platform’s own access logs
    • · All data the application processes, because the application runs there

    Railway operates the servers and holds platform-level logs under its own terms; we have not verified a retention period for those logs against a signed agreement and do not state one here. The application’s own records keep no full IP addresses — see the retention table.

  • PostgreSQL on Railway

    Storing account, subscription, device, usage-ledger and billing-event records. Same vendor and same us-east4 region as the application server.

    • · All persisted application data

    A PostgreSQL instance on a Railway volume, reachable only on Railway’s private network and never exposed to the public internet. It is not a separate managed-database vendor: the data stays with the hosting provider named above, under the same terms.

  • Resend

    Terms not verified

    Sending transactional email: address verification, sign-in links, security notices and billing notifications.

    • · Your email address
    • · The content of the transactional message itself

    Resend is the application’s outbound mail provider. How long it keeps message content and delivery logs is governed by its own terms, which we have not verified against a signed agreement. Receiving mail is a separate job: the mailboxes named on this site are hosted by the domain’s inbound mail provider, and Resend is not involved in them.

  • Cloudflare R2

    Off-site storage of encrypted logical database backups, held with a different vendor from the database itself.

    • · An encrypted archive of the application database: account, subscription, device, usage-ledger and billing-event rows
    • · No audio, transcripts, meeting recordings or Projects — those never reach this server

    Each archive is encrypted with AES-256-GCM on our own server before it is uploaded, so Cloudflare holds ciphertext and never holds the key. Off-site backups are switched off on the current deployment and no archive exists yet; this row is here because R2 is where they will go the moment they are switched on.

How long things are kept

Audio on Xuula servers
Not stored
Transcripts on Xuula servers
Not stored by default
Meeting recordings, transcripts and Projects
On your PC, for as long as you keep them
Usage ledger (counts only)
24 months
Account records after deletion
30 days

What you control

  • Turning AI cleanup off, so no transcript text is sent for rewriting.
  • AI meeting insights run only when you ask for them — there is no background processing.
  • Muting your microphone during a meeting recording, and pausing or stopping the recording at any time.
  • Turning live translation on or off per meeting.
  • Clearing your local dictation history at any time from the app; meeting files are ordinary local files you can delete yourself.
  • Exporting everything your account holds, as JSON.
  • Deleting your account, which removes your data after a short recovery window.

Recording a conversation can require the notice or consent of the people in it, depending on the jurisdictions involved. You are responsible for obtaining any legally required consent before recording a meeting.

Read the full Privacy Policy