Privacy Policy

Effective and last updated: September 4, 2026

saidtrack is a realtime meeting copilot operated by Prinsur Tech CO., LTD. (鈦溪科技股份有限公司), a company incorporated in Taiwan ("saidtrack", "we", "us"). This policy explains how we collect, use, store, and share personal data when you visit saidtrack.com or use the saidtrack service. It applies to account holders and also to meeting participants who are recorded during a meeting without holding an account. The English version of this policy controls; the Traditional Chinese version is provided for convenience.

1. Who is responsible for your data

Prinsur Tech CO., LTD. is the data controller for the personal data described in this policy. You can reach us at support@prinsur.com for any privacy question or request.

2. What we collect

Account data. Your name, email address, and optional profile picture, received through our identity provider WorkOS when you sign in with a one-time email code or with Google or GitHub. The Service uses no passwords. We also store your display preferences on your account: the language you read the Service in, and the time zone we show your dates and times in.
Meeting audio. The live audio of meetings you record. Meetings are conversations, so the audio can include the voices of people other than you.
Transcripts and speaker labels. The text transcript of each meeting, segmented into turns and attributed to you or to other speakers.
Voiceprint (biometric data). If you choose to enroll, we process a short sample of your own voice, live and without storing it, and compute a numeric voice embedding from it. The embedding is what we keep; see section 5.
Meeting inputs. The meeting objective you enter, and the text extracted in your browser from files you choose to attach as meeting knowledge. The files themselves never leave your browser; only the extracted text is sent to us.
AI outputs and inferences. Summaries, suggestion cards, and emotion signals that our systems generate from the transcript.
Device and usage records. Our identity provider records each sign-in session so you can review it and sign it out from your settings. Alongside each one we store the sign-in method, the name of the browser and operating system you used, and an approximate location no more precise than a city. We do not store your IP address: our hosting platform resolves the location at its edge, and only the resolved city and country reach us. We delete these details when you sign that session out, and the next time you open your security settings we delete them for every session that is no longer active. Deleting your account removes them as well. Our servers keep operational logs for security and debugging. When the service hits an error, technical data about that error is sent to our error reporting provider; it carries no meeting audio, transcripts, or AI message content.

3. Cookies and analytics

We use first-party cookies only. There are no third-party advertising cookies and no cross-site tracking.

Session cookie (wos-session): keeps you signed in. Set by our authentication layer, httpOnly, for up to 400 days, and deleted when you sign out.
Locale cookie (saidtrack_locale): remembers your language preference for up to one year.
Time zone cookie (saidtrack_timezone): remembers the time zone we show your dates and times in for up to one hour, and is set again on your next visit after it expires. It comes from your browser's own time zone setting; we never derive it from your IP address.
Pending verification cookie (workos_pending_verification): set for up to 15 minutes while a sign-in needs its email address verified; contains a short-lived verification token and the email address being verified.
Interface cookies: remember sidebar state for up to 7 days. Your theme preference is kept in your browser's local storage.

We use Vercel Web Analytics and Vercel Speed Insights to understand aggregate page usage and performance across the site, including for visitors without an account. These tools do not set cookies and do not build cross-site profiles.

4. Meeting participants who are not account holders

saidtrack transcribes meetings, and meetings include people other than the account holder. If you were a participant in a recorded meeting but do not hold a saidtrack account, this section is for you.

The account holder who records a meeting is responsible for informing participants and obtaining any consent required by applicable law, as set out in our Terms of Service.
What we process about participants: their voice within the meeting audio (processed live and not stored), their words within the transcript, and speaker labels that distinguish them from the account holder.
How remote participants' voices reach us. When a meeting is held online, the account holder can share the browser tab the meeting runs in, and the audio of that tab, which carries the voices of everyone speaking in the meeting, is sent to us together with their microphone. Nothing is installed in the meeting, saidtrack does not join it, and it does not appear in the participant list, so this is not visible to you from inside the meeting. The account holder remains responsible for telling you.
We do not create voiceprints of participants, and we do not build participant profiles across meetings. Voiceprint enrollment exists only for the account holder's own voice. To decide which parts of a meeting were spoken by the account holder, our system computes a numeric vector from short stretches of the live audio and compares it against the account holder's own enrolled voiceprint. That vector is discarded immediately after the comparison: it is never stored, never written to logs, never used to identify anyone, and never used to build a profile of any other person.
Reports published as links. The account holder can publish the report of a meeting as a link that anyone holding it can open, as described in section 9. We do not notify you when this happens, and it is not visible to you from inside the meeting. The account holder remains responsible for telling you.
You can contact us at support@prinsur.com to exercise your rights. We can delete a meeting in full and correct speaker labels. We do not currently have an automated tool to remove a single speaker's words from a transcript, so such requests are handled manually within a reasonable period. Deleting a meeting also destroys every link that published it. Correcting a speaker label changes the report in the product, but it does not change a report that has already been published as a link, so on request we will also withdraw an affected link.

5. Voiceprints and biometric data

A voiceprint is a numeric representation of your voice. It counts as biometric data in many jurisdictions, so we hold it to the strictest rules we apply to any data category.

Optional and self-only. Enrollment is optional, and the service only supports enrolling your own voice. Enrolling another person's voice is prohibited by the Terms of Service.
Single purpose. The voice embedding is used for exactly one thing: recognizing which parts of a meeting were spoken by you, so your transcript can be labeled correctly. The voice sample you record is used only to compute that embedding and to mark where you are in the text while you read. Neither is used for identification across services, advertising, or any other purpose.
Separate, recorded consent. Before any voice sample is processed, you must give explicit consent through a dedicated consent step, and we store the timestamp of that consent. For users in the European Economic Area this is the explicit consent basis under Article 9(2)(a) GDPR.
The enrollment voice sample. While you read the enrollment prompt aloud, that audio is streamed to Soniox, our speech-to-text provider, solely to mark where you are in the text on screen. We do not store it and do not write it to logs, and the connection is torn down when the segment ends; Soniox's own handling is described in section 7. The embedding itself is computed inside our own service and is never sent to an external AI provider.
Storage and sharing. The voiceprint is stored as a numeric voice embedding in our database, encrypted at rest. The embedding is never returned through any API, never shared with our speech-to-text provider or our LLM provider, and never leaves our systems.
Destruction. Your voiceprint is destroyed when you withdraw consent in settings or when you delete your account, whichever comes first.

6. How we use personal data

We use the data described above to provide realtime transcription and in-meeting suggestions, to generate summaries and reports after a meeting, to attribute speech to speakers, to manage your account, to keep the service secure, to debug the service using error and diagnostic records, and to improve it using aggregate, non-identifying information.

We do not sell personal data.
We do not use your data for advertising.
We do not train AI models on your recordings, transcripts, or files, and our AI providers are, by contract or by their default policies, not permitted to do so either.

7. AI processing and our providers

Speech-to-text: Soniox. Meeting audio is streamed to Soniox for live transcription and is processed in real time. We do not store audio (see section 11). According to Soniox's public security statements, realtime audio and transcripts are not stored by Soniox and are not used to train its models. We rely on those statements and on our agreement with Soniox. If you enroll a voiceprint, the speech you read aloud is streamed to Soniox as well, solely to mark where you are in the text on screen; see section 5.
Language model: Anthropic. Transcript text, your meeting objective, and any attached knowledge text are sent to Anthropic's Claude models to generate summaries, suggestions, and emotion signals. Anthropic does not train its models on commercial API inputs or outputs by default. Under Anthropic's standard data retention policy, API inputs and outputs are retained by Anthropic for up to 30 days and then deleted.
Emotion signals are text inferences. Emotion signals shown in the product are inferences drawn by a language model from the text transcript. We do not analyze voice tone or any biometric signal to detect emotion.

8. Service providers

Apart from a report an account holder chooses to publish as a link (see section 9), we share personal data only with the service providers that run the product for us, and only to the extent needed for their role. The providers central to the product are WorkOS (identity and sign-in), Soniox (speech-to-text), and Anthropic (language model). The full list, including hosting and storage providers, is maintained on our Subprocessors page.

An account holder can publish the report of a finished meeting as a link that opens without signing in. The product calls this sharing a report. Anyone holding the link can read it, including anyone it is forwarded to and the link preview and security scanning services that chat tools and mail systems run automatically. The unguessable address is the only thing protecting a published report: there is no password, and we do not check who the reader is. Published reports are marked not to be indexed and are kept out of our sitemap, but that is a request to search engines rather than a barrier.

What a link publishes. Only the parts the account holder selects: the summary, the transcript, the emotion signals, the suggestion cards. A link also carries the meeting title, the account holder's display name, and the dates the report was generated and the link was published. It never carries the account holder's email address or profile picture, a voiceprint, or any internal identifier, and participants other than the account holder appear only as generic speaker labels. The summary, the suggestion cards, and the emotion signals can all restate what was said in the meeting, so publishing any of them can expose meeting speech even when the transcript itself is not published, and a suggestion card can also reflect knowledge text the account holder attached to the meeting. Anything said out loud during the meeting, including names, email addresses and numbers, appears in whatever parts are published, like any other words.
A published report is frozen. A link shows the report as it stood when the link was created. Regenerating a summary, re-running the emotion analysis, or correcting a speaker label changes the report in the product, but it does not change what an already-published link shows. Publishing a new link is the only way to share a corrected report.
How a link ends. Every link is created either with an expiry of 7, 30, or 90 days, or with no expiry at all. A link stops working when its expiry passes, when the account holder asks us to withdraw it, and when the meeting or the account is deleted, which takes every link to that meeting with it. A link can also stop working if we retire the format it was published in; we treat that the same as an expiry, and a new link has to be published. Withdrawal and expiry take effect for everyone the next time the link is fetched from us.
What ending a link cannot do. It cannot reach a report a reader has already opened, copied, or saved. Because a published report opens without an account, the product does not show who opened a link or how many times.

We do not notify meeting participants when a report is published, and it is not visible to them from inside the meeting. Telling them is the account holder's responsibility, as set out in our Terms of Service. If you publish a link, send it only to people you would be willing to show the whole report to.

10. International transfers

We operate from Taiwan. Persistent data (accounts, transcripts, summaries, voiceprints) is stored with database and compute providers in Singapore. Short-lived in-meeting state is held in a globally replicated cache for the duration of a meeting. Our speech-to-text, language model, identity, email delivery, web hosting, profile picture storage, and error reporting providers process data in the United States. Our status page and availability monitoring provider is a United States company that, per its own privacy policy, processes personal data primarily in the European Union. Where the GDPR applies, transfers rely on recognized mechanisms such as the European Commission's Standard Contractual Clauses, implemented through data processing agreements with our providers. We handle cross-border transfers in accordance with Taiwan's Personal Data Protection Act.

11. How long we keep data

Meeting audio: not stored. Audio is processed in memory in a rolling buffer of roughly 90 seconds and immediately discarded. It is never written to disk or object storage, on our systems or at our direction.
Voiceprint enrollment voice sample: not stored. It is processed live and discarded, and never written to disk or logs.
Transcripts, summaries, suggestions, emotion signals: kept until you delete the meeting or your account.
Published report copies: each published link holds its own frozen copy of what it published. A copy is deleted when you delete the meeting or your account, and when our systems next run housekeeping after the link's expiry has passed. Housekeeping is not on a fixed schedule, so an expired copy can persist for a while after the link stops working; it is unreachable by anyone from the moment the expiry passes. Withdrawing a link stops it opening but does not delete its copy, which we keep as the record of what was published.
Voiceprint: kept until you withdraw consent or delete your account.
Meeting knowledge and live meeting state: held in a memory cache during the meeting, actively cleared when the meeting ends, and in any case expiring within 6 hours.
Audit records: audit records tied to a meeting are deleted together with that meeting or your account. A small set of system-level audit records is retained for compliance and security. This includes the timestamp of your voiceprint consent, which we keep after withdrawal as evidence that consent existed and was revoked.
Account data: kept until you delete your account.
Server logs and error reports: retained for a limited period by our hosting platforms and our error reporting provider. Log contents are systematically masked and do not include transcript text.
Database change history: 6 hours. Our database provider keeps a six-hour change history for point-in-time restore. Content you delete stays in that history for those six hours and then rolls out of it. We do not access it for any purpose other than restoring the database after an incident.

Deleting your account permanently deletes your account, meetings, transcripts, summaries, published report copies, voiceprint, and profile picture, and clears any live meeting state, subject only to the audit exception and the six-hour change history above.

12. Security

Data is encrypted in transit with TLS and encrypted at rest at the infrastructure layer. Internal service endpoints are protected by shared secrets, and logs are masked to keep transcript content out of them. Our providers maintain independent certifications such as SOC 2 and ISO 27001. No method of transmission or storage is completely secure, but we design the service so that the most sensitive data, your audio, is never stored at all.

13. Your rights

Depending on where you live, you may have the right to access, correct, delete, restrict, or object to the processing of your personal data, and the right to data portability. Residents of Taiwan have the rights set out in Article 3 of the Personal Data Protection Act, which cannot be waived in advance. Meeting participants without an account can exercise these rights too.

In the product you can delete meetings, delete your account, withdraw voiceprint consent, and export any meeting report as a DOCX or Markdown file. For anything else, or to exercise rights by request, email support@prinsur.com. We respond within a reasonable period and may need to verify your identity before acting on a request.

Sign-in with Google. When you choose to sign in with Google, we receive only your basic profile information: your name, email address, and profile picture, through Google's standard identity scopes (openid, email, profile). We use this information solely to create and authenticate your saidtrack account and to display your identity within the product. We do not request or access your Gmail, Google Calendar, Google Drive, or any other Google content. saidtrack's use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements.

Sign-in with GitHub. When you sign in with GitHub, we receive only your basic GitHub profile and email address, used for the same account purposes.

Both sign-in flows are brokered by WorkOS, our identity provider, which processes this profile data on our behalf.

15. Children

The service is not directed to children under 16. By using saidtrack you confirm that you are at least 16 years old. If we learn that we have collected personal data from a child under 16, we will delete it.

16. Changes to this policy

We will post any changes on this page and update the date at the top. For material changes, we will provide notice in the product before the change takes effect.

17. Contact

Prinsur Tech CO., LTD. (鈦溪科技股份有限公司), Taiwan
support@prinsur.com