Skip to main content
← Back to Transkripsie

Sub-processors

Last updated: September 25, 2026

To run Transkripsie, we use third-party service providers (“sub-processors”) that may process personal data for the features you request. The list below describes intended data flows; the contractual status, retention terms, and international-transfer safeguards for each provider must be confirmed before the relevant processing begins.

We aim to give advance notice before adding or replacing a sub-processor that processes personal data. To subscribe to change notifications, email privacy@transkripsie.com with the subject line “Subscribe: sub-processor updates”.

Current sub-processors

Sub-processorPurposeAI purposeLocationData shared
Hetzner Online GmbHApplication server, database and recordings on server disk and block-storage volume—Germany (Falkenstein)Account data and stored service content handled by the configured hosting and storage environment; encryption coverage depends on production configuration
Recall.ai, Inc.Meeting bot infrastructure for Google Meet, Zoom, Microsoft Teams, and WebexCaptures and relays live meeting audio/video for downstream speech recognition (does not generate AI output)USAMeeting join URL, bot session metadata, and the live meeting audio/video stream during recording
Soniox, Inc.Speech-to-text transcription (batch uploads and live streaming)Automated speech recognition and speaker diarisationUSAAudio sent for transcription. Soniox storage and deletion depend on the API service used and the deletion workflow; we do not promise a fixed processing-window deletion period
OpenRouter, Inc.Primary generative-AI routing layerSummaries, chapter headings, meeting titles, chat responses, and translationUSATranscript text submitted for requested AI processing. Our requests ask OpenRouter to use only providers that do not keep prompts or collect them for training; retention and model-training treatment otherwise depend on the provider policy and any executed agreement
Microsoft Azure (via OpenRouter)Runs the default language model for generative-AI features (OpenAI's GPT-5 mini)Summaries, chapter headings, meeting titles, chat responses, and translationSweden (Azure Sweden Central) or another Azure region OpenRouter selectsTranscript text relayed by OpenRouter for requested AI processing; retention and model-training treatment depend on the provider policy and any executed agreement
xAI Corp.Fallback generative-AI providerSummaries, meeting titles, chat responses, and translation, used directly only if OpenRouter is not configured or we explicitly switch on a fallback to providers that may retain dataUSATranscript text submitted for requested AI processing; retention and model-training treatment depend on the provider policy and any executed agreement
Anthropic, PBCFallback generative-AI providerSummaries, meeting titles, chat responses, and translation, used directly (after xAI) only if OpenRouter is not configured or we explicitly switch on a fallback to providers that may retain dataUSATranscript text submitted for requested AI processing; retention and model-training treatment depend on the provider policy and any executed agreement
Zoho CorporationTransactional email delivery (sign-up, password reset, security alerts, receipts)—USA (Zoho US data centre)Recipient email address, name, and message content
Google LLCGoogle Calendar integration (only for users who connect a Google account)—USAOAuth tokens and calendar event metadata (title, time, attendees, conferencing link)
Backblaze, Inc.Encrypted off-site backups (B2 object storage)—USAEncrypted database and audio backup artifacts when a backup run succeeds. Backup key custody, immutability, lifecycle retention, alerting, and complete media recovery are under operational review
Stripe, Inc.Payment processing (when billing is enabled on your account)—USAName, email, billing address, and payment method tokens. Full card numbers are not transmitted to or stored by Transkripsie
Functional Software, Inc. (Sentry)Application error monitoring and performance tracing, when enabled—USAStack traces, request paths, and diagnostic metadata when enabled; production enablement and scrubbing coverage must be confirmed
Healthchecks.ioBackup and restore-drill job monitoring, when configured—GermanyOperational ping metadata only when configured; current production alerting coverage is under review
PostHog, Inc. (US Cloud)Optional product analytics and session replay on public website pages — only if enabled by us and the visitor opts in to analytics—United States (US Cloud)Pages viewed, clicks, and session replays of public pages with all form inputs masked; random analytics identifiers. Visitor IP addresses are not forwarded (requests are relayed through our own domain). Never used inside the app, on recordings, or on shared links
ThreadchatOptional RAG integration (only for users who link a Threadchat account)Optional transcript indexing for retrieval-augmented search (user-initiated integration only)Varies (per-user VPS)Transcript text, summaries, and metadata pushed for search/RAG indexing when the integration is enabled

Notes

  • The speech recognition provider is Soniox, Inc. (USA). Soniox’s published materials say that real-time requests are transient and that stored asynchronous content can be deleted through the API; asynchronous file storage is service-dependent and may require manual deletion. Our integration must complete and verify that cleanup before we promise a fixed retention period.
  • Generative-AI features are routed primarily through OpenRouter, Inc., where the default model is OpenAI’s GPT-5 mini. Because we ask OpenRouter to use only providers that do not keep prompts, OpenRouter runs that model on Microsoft Azure, in Azure Sweden Central or another Azure region it selects. xAI Corp. and Anthropic, PBC are backup providers. We call the backup providers directly only if OpenRouter is not configured or we explicitly switch on a fallback to providers that may retain data; otherwise a failed OpenRouter request is not re-sent to them. Processing, retention, and model-training restrictions depend on each provider’s current policy and any executed agreement; we do not represent that these providers have identical contractual restrictions.
  • Stripe is engaged only when billing is enabled on your account. If you have a free account, no data is shared with Stripe.
  • Google is engaged only if you choose to connect Google Calendar. If you do not connect a Google account, no data is shared with Google through Transkripsie.
  • Backup artifacts sent to Backblaze may be encrypted before upload. Object Lock, lifecycle retention, failure alerting, key custody, and complete audio restore coverage are under operational review, so we do not represent the current backup service as immutable or as having a fixed deletion window.
  • If Sentry is enabled, it is intended to receive diagnostic metadata rather than audio or transcript content; the production enablement and scrubbing configuration must be confirmed.
  • PostHog is engaged only when we enable it and a visitor opts in to analytics in our cookie preferences. Our own first-party visit statistics (cookieless for every visitor, with a persistent visitor ID only after you opt in to analytics) run on our Hetzner servers and involve no additional sub-processor.
  • Approximate visitor location is looked up on our own servers from the DB-IP “IP to City Lite” database, which we download; DB-IP is a data source, not a sub-processor, and receives no personal data. IP Geolocation by DB-IP.
  • Threadchat is engaged only if you explicitly link a Threadchat account for RAG search. Otherwise no data is shared with Threadchat.
  • Enterprise single sign-on through WorkOS, Inc. is not enabled. We will add WorkOS to this list before enabling it.

Related

Last updated September 25, 2026. To subscribe to changes, email privacy@transkripsie.com.