Legal
Privacy Policy
Last updated: August 4, 2026
1. Who we are and how to contact us
TCVoiceAI is an independent desktop voice translation product operated by its creator. The operator of TCVoiceAI is the controller for account, website, billing, support, and service-usage data processed for this service. Privacy questions and data-rights requests can be sent to support@tcvoiceai.com.
If your jurisdiction requires additional controller, representative, or data protection officer details, contact us and we will provide the currently applicable contact information for your request.
2. What we collect
When you sign in, we collect Google account information needed to create and secure your account: name, email address, verified email status, Google account identifier, avatar URL, and related authentication records. We also store account role, locale if provided, referral information, support preferences, plan and subscription state, Paddle customer/subscription identifiers, invoices and transaction references made available by Paddle, and billing status needed to provide paid plans.
For service operation, we store request identifiers, language direction, shared translated and suggested-reply word usage, summary word counts, quota reservations, daily usage counters, OpenAI model identifiers, token counts, provider request/response identifiers, timestamps, and other metadata needed to enforce plan limits, prevent abuse, troubleshoot reliability, and calculate costs. This metadata is designed not to include raw audio, transcript text, translated text, prompt context, generated reply text, or generated summary content.
Automated abuse-prevention controls evaluate service-usage metadata such as request counts and timing, request type, estimated token and provider usage, quota state, and provider attempts. These controls may temporarily restrict AI features. Longer AI access suspensions can be reviewed by Support on request. Stored abuse-prevention records are designed not to include audio, transcripts, translations, prompt context, or generated content.
If you explicitly connect an optional work integration, we store its provider key, account label, provider-verified scopes, token expiry, revocation state, and encrypted access and refresh credentials needed to perform the actions you approve. Connector credentials are never returned by the API. Each publication accepts only the reviewed task, note, email-draft, or calendar-event fields that you explicitly approve; raw transcripts are rejected.
If Google Analytics is enabled and you choose Allow analytics, Google may receive and process your IP address to derive an approximate region, together with a page URL without its query string, page title, referrer without its query string, browser and device information, and basic visit, interaction, and session statistics. Google Analytics uses first-party analytics identifiers to distinguish visits. Our integration does not attach your TCVoiceAI account ID, name, email address, audio, transcripts, translations, summaries, or support content to analytics events.
If you contact us, create a support ticket, or send app feedback or a review, we process the name, email address, subject, category, priority, rating, message content, replies, public review consent, administrative notes, and timestamps needed to respond and moderate submissions. Public reviews appear on the site only when you opt in and an admin approves them. If the macOS app sends feedback or an error report, we may receive app version, build, macOS version, architecture, audio backend, selected device names, app state, quota counters, plan key, request identifiers, error category, error message, and diagnostic metadata. Feedback and reports are intended to exclude audio, transcripts, credentials, and conversation content.
Support tickets may include the app platform selected in the form.
3. Voice, transcript, translation, suggestion, and AI output content
After you actively start Translated voice or Original audio, the macOS app captures audio from your selected microphone and the TCVoiceAI virtual speaker route. Apple Speech Recognition converts speech to text. Translated voice also uses macOS text-to-speech for spoken translation. Original audio keeps real voices live and can create bilingual History or same-language transcription, with optional transcript translation.
Our server receives recognized text, language direction, bounded conversation context, optional user context notes, and only matching custom vocabulary terms from the active Conversation Profile when a translation is requested. A same-language transcription-only session does not send translation requests. Profile names, contact labels, and the complete profile library are not sent to TCVoiceAI servers. While a session is active, terms for the recognized language may be supplied to Apple Speech Recognition as contextual hints under your macOS settings and Apple’s privacy terms. Matching vocabulary sent for translation is processed with that request and is designed not to be stored afterward. When you explicitly request AI conversation output, the app sends original speaker turns from local History with your optional request to create the requested result.
If you enable Suggested replies during a bilingual or same-language session, the app may ask for a suggestion only after the partner completes a sufficiently substantive turn. In bilingual sessions, routing can reuse the current translation response. In same-language sessions, a separate routing request is made without requesting translation. Suggestion requests may include the completed partner turn, up to four recent original turns, your bounded context note, and up to eight matching vocabulary pairs. Simple greetings, acknowledgements, and basic personal questions are filtered out.
Research-routed suggestions may optionally let OpenAI use Web Search when current public information could improve the reply. If search is used, source links are shown separately in the app; they are not inserted into the suggested reply or stored by TCVoiceAI.
Translation, suggested-reply, and v1 plain-summary requests are designed not to retain raw audio, transcript text, translation text, prompt context, generated reply text, or generated summary content after the response is returned. OpenAI requests use storage disabled (store: false). We retain usage metadata such as word counts, token counts, model identifiers, and request IDs.
When the default-disabled structured Outcome recovery API is enabled, the full source transcript and optional prompt are not written as standalone records in the TCVoiceAI application database. The generated Outcome response is stored as application-encrypted replay ciphertext for up to 24 hours by default so the app can recover a lost response. That generated response can include evidence excerpts of up to 500 Unicode characters for each cited transcript turn. The ciphertext is deleted early when the app acknowledges durable local saving, or by scheduled expiry after its replay period. A content-free status tombstone may be retained for 30 days by default for idempotency and reliability handling.
The connector broker is disabled by default. When you enable a supported integration and approve a publication, TCVoiceAI sends only that approved work-result payload to the destination provider you selected. A retryable payload is application-encrypted for up to seven days by default and is removed immediately after confirmed delivery, terminal failure, cancellation, or expiry. Content-free publication receipts may remain for 30 days by default. If delivery races connection revocation, the receipt is marked delivery ambiguous rather than claiming that the remote item was cancelled.
Background helper privacy: while the main app is idle or closed, the helper can relay audio buffers between your selected physical and virtual devices so normal calls continue without repeatedly changing audio settings. It handles those buffers only for routing and does not perform speech recognition, transcription, translation, text-to-speech, recording, storage, content inspection or analysis, upload, or server communication. The helper is not used to listen to or collect your conversations.
Conversation Profiles, per-profile History, exported transcript files, and exported settings files are controlled by the app on your device rather than stored in the TCVoiceAI server database.
The export includes profile languages, modes, context notes, vocabulary, the active profile, interface language, and compatible History, Floating captions, and suggested-reply preferences.
It does not include sign-in tokens, passwords, audio-device choices, diagnostics consent, conversation History, summaries, or other sensitive or device-specific data.
Keep exported settings files private because profile names, context notes, and custom vocabulary may contain personal or confidential information.
Floating captions reuse recent local History and the current suggested reply on your Mac. Showing the overlay does not create another audio capture, recognition task, translation request, or server-side transcript. The app asks macOS to exclude the overlay from normal window capture where supported, but you should hide it before sharing or recording an entire display that contains sensitive conversation text.
4. Why we process data
We process account, authentication, entitlement, translation, suggested reply, summary, optional connector, quota, billing, and checkout data to perform our contract with you and provide the service you request. Connector authorization and every connector publication additionally require an explicit in-app action. We process security, abuse-prevention, logging, diagnostic, and support data for our legitimate interests in keeping the service reliable and safe. We process invoice, tax, accounting, and payment records where needed to comply with legal obligations. Optional Google Analytics processing is based on your consent. You may refuse it without losing website features and withdraw it at any time through Cookie settings in the footer.
5. Service providers and international transfers
We use providers for authentication, payments, hosting/infrastructure, email delivery, speech and AI processing, support, logging, reliability, analytics, and abuse prevention. Current core providers include Google for sign-in and, only when enabled and consented to, Google Analytics; Paddle for checkout, subscription, tax, and payment processing; OpenAI for translation, suggested replies, and AI summaries; Apple platform services used by the macOS app; and Resend for service emails where configured.
Google describes how it handles information in the Google Privacy Policy. Your Analytics choice controls whether our public pages load the Google Analytics tag; it does not change the separate Google sign-in processing you request when you choose to sign in.
These providers may process data in countries other than your own. Where required, we rely on appropriate safeguards such as provider data processing terms, standard contractual clauses, adequacy decisions, or other lawful transfer mechanisms.
Optional connector providers are not contacted until you choose a supported integration. Their own privacy terms apply to the account credentials, destination identifiers, and approved work-result content that you direct TCVoiceAI to send to them. The app must show the provider and destination before each publication; TCVoiceAI does not publish an entire transcript through this interface.
7. Retention and security
We retain account, billing, entitlement, usage metadata, support, contact, referral, diagnostic, and administrative records for as long as needed to provide the service, comply with tax/accounting and payment obligations, resolve disputes, enforce terms, maintain security, and prevent abuse. Translation, suggested-reply, and v1 plain-summary content is designed not to be retained after the response is returned. If structured Outcome recovery is enabled, the generated Outcome, including bounded evidence excerpts for cited turns, has temporary encrypted replay retention for up to 24 hours by default; an acknowledgement deletes it early, while a content-free status tombstone may remain for 30 days by default. Connector retry payloads are encrypted for up to seven days by default and connector receipts for 30 days. On disconnect, local publishing stops immediately; encrypted credentials may be retained for up to 24 hours by default solely for bounded remote-revocation retries, then are deleted. A content-free provider/status/error audit may be retained with an account-deletion record when remote revocation could not be confirmed before deletion.
We use reasonable technical and organizational safeguards, including authentication, role-based admin access, CSRF protection, throttling, usage limits, diagnostic filtering, and provider security controls. No internet service can be guaranteed to be perfectly secure.
8. Your privacy rights
Depending on where you live, you may have rights to request access, correction, deletion, restriction, portability, objection to certain processing, withdrawal of consent, or review of automated decisions. EU/EEA and UK users can read more on the GDPR rights page.
Some data may need to be retained or excluded from deletion where required for billing, tax, fraud prevention, security, legal claims, or compliance. We may need to verify your identity before fulfilling a request.
9. Children and sensitive use
TCVoiceAI is not directed to children, and paid accounts should be used only by people who can lawfully create an account and subscribe to software services. Do not use the service to process special-category, highly sensitive, or regulated information unless you have confirmed that the service is appropriate for that use and you have all required rights and consents.
10. Changes
We may update this policy as the product, providers, or legal requirements change. If changes are material, we will make reasonable efforts to notify users through the website, app, email, or account notices.