Tymbr
Privacy policy
Tymbr turns your recordings into transcripts and can import existing Google Meet transcripts when you connect Google Calendar and Meet. Recordings, transcripts and participant information are personal information, and sometimes special personal information. This policy explains what we do with them, in plain terms, as the Protection of Personal Information Act (POPIA) requires.
1. Who is responsible
Tymbr is operated by Gareth Saul trading as 20 East, a sole proprietor in South Africa. For your account information, we are the responsible party. Our Information Officer is Gareth Saul, reachable at hello@tymbr.dev or 18 Lyndhurst Road, Kenilworth, Cape Town, 7708. We do not have a telephone line. Email is the way to reach us with questions and complaints.
2. Your recordings: two roles
Your recordings contain other people's voices and words. For that content, you are the responsible party under POPIA and we are your operator. We do not decide what is recorded or why. Our terms of service are the written operator agreement that section 21(2) of POPIA requires. In practice this means:
- You decide what to record and must have a lawful basis to do so. Section 4 of the Regulation of Interception of Communications Act lets a party to a conversation record it. Telling participants where the law or your professional rules require it is your responsibility.
- People in your recordings who want to see, correct or delete their information must direct that request to you, not to us. If such a request reaches us, we will forward it to you within five business days and leave the decision with you. Tymbr gives you the tools to act on it.
- We process the content only on your instructions. Uploading a recording is an instruction to transcribe it, and deleting one is an instruction we carry out without question.
- We keep the content confidential, secure it as section 8 describes, and tell you as soon as we reasonably can if we believe it has been accessed without authority.
- You must obtain a person's explicit consent before you enrol their voiceprint, keep evidence of it, and delete the voiceprint if they withdraw it. See section 3.
Our staff do not read or listen to your content except with your documented permission for specific support work, when necessary to investigate security incidents or abuse, or where the law requires it. Automated processing runs the features described below. We do not check whether you had the right to make a recording. We do not use your content to train or improve general AI models or the underlying voice-recognition model.
3. What we collect and why
| Information | Why we process it | Lawful basis |
|---|---|---|
| Your email address and sign-in identity | To create your account, keep your data separate from everyone else's, and contact you about the service | Performing our contract with you |
| Recordings you upload, record or dictate | To produce transcripts and let you play back audio | Performing our contract with you; for other people in the recording, you are the responsible party |
| Transcripts and your edits | To show, search and export your transcripts | Performing our contract with you |
| Connected Google calendar events: titles, times, attendees, email addresses and meeting links | To organise meetings, match recordings to calls and supply context for automatic meeting titles | Your separate permission to connect Google; performing the features you request |
| Existing Google Meet transcripts, speech timestamps, participant names and Google identifiers | To show, search and export transcripts, match calls to recordings, identify speakers and update private recognition profiles for people already remembered in your account | Your separate permission to connect Google; you must have the rights and permissions needed for the meeting content and any enrolled voiceprints |
| Speaker rosters (names you assign to speakers) | To label who said what | Performing our contract with you |
| Voiceprints, if you enrol them | To recognise the same speaker across recordings | Consent, under section 27(1)(a) of POPIA. Voiceprints are biometric information and are special personal information under section 26. You can withdraw consent by deleting the voiceprint. |
| Device credentials for the desktop recorder | To let the recorder upload to your account and let you revoke it | Performing our contract with you |
| Purchases and credit usage | To sell credit, deduct it per minute, show your balance and meet tax and accounting duties | Performing our contract with you; legal obligation |
| Technical logs (IP address, request times, errors) | To keep the service secure and working | Our legitimate interest in security and reliability |
We collect only the information described here. There is no tracking for advertising, and we do not buy personal information or collect it from unrelated sources.
Google Calendar and Meet: permission and use
Google sign-in requests your identity information. Calendar and Meet access is optional and granted separately. Before you connect or reconnect an identity, or add another Google account, Tymbr shows a notice explaining imports, retention and use, followed by Google's permission screen. You can cancel instead of granting access.
After you connect, Tymbr reads recent and upcoming events from that identity's primary calendar and imports existing Meet transcripts that Google permits it to access. Sync runs in the background, including while you are signed out. The connection does not start recording or transcription in Meet and does not read the contents of Gemini smart-note documents.
To generate a title for your own recording, Tymbr sends its transcript and matching calendar context to OpenAI through Cloudflare AI Gateway. Calendar context can include the event name, whether it is recurring, organiser email, attendee names and email addresses, and your account's email domains. Cloudflare Workers AI provides a fallback title model. This processing happens on those services' infrastructure, rather than only on your device or our own server. We disable AI Gateway request logging for these title requests; this is separate from each provider's retention controls.
Meet data can name speakers in your own recordings and associate names and Google identifiers with people in your account. For people you have already chosen to remember, those names can help associate samples from your own recordings with their private voice-recognition profiles. Appearing in Meet alone does not enrol someone as a remembered voice. Profiles stay within your account; we do not use them to train or improve the underlying recognition model.
Voice recognition uses a self-hosted SpeechBrain model on infrastructure we control at Fly.io. Its model files are downloaded when we build the service and loaded offline during inference. Your audio, Google data and private recognition profiles are not sent to SpeechBrain or Hugging Face for training or secondary purposes. Processing is on our hosted infrastructure, not on your device.
Tymbr's use and transfer of information received from Google APIs will adhere to the Google API Services User Data Policy, including its Limited Use requirements, and the Google Workspace user data and developer policy. These restrictions also cover information derived from Google data. We use it only for the features you request. We do not sell it, use it for advertising, credit scoring or lending decisions, or use it to train or improve general AI models.
Raw, aggregated and derived data received from Google Workspace APIs is not used to develop, improve or train foundational or generalised artificial intelligence or machine-learning models. Third-party processing of that data is limited to providing the user-facing features described here and must comply with Google's Limited Use requirements.
For transcription and recording titles, we use commercial Mistral and OpenAI APIs, rather than their consumer chat services. We use providers only under terms and account settings that exclude your content from general model training, and do not opt your content into model-training data sharing. The Cloudflare Workers AI fallback is also limited to inference. Maintaining a private voice-recognition profile does not train or change the underlying SpeechBrain model.
4. Children
Tymbr is for adults and businesses. We do not knowingly open accounts for anyone under 18. If a child's voice appears in your recording, you are responsible for having the consent of a parent or guardian where POPIA requires it.
5. Who we share it with
We do not sell or rent your information. We share it only with operators that we need to run the service, under written terms that bind them to confidentiality and security:
| Operator | What they do | What they receive |
|---|---|---|
| Cloudflare | Hosts the web app, database, file storage and job queue; routes title requests through AI Gateway and runs the fallback title model in Workers AI | Account data, recordings and transcripts for hosting; recording transcripts and matching calendar context for title generation |
| OpenAI | Generates automatic meeting titles through its API | Your recording's transcript and matching Google calendar context, through Cloudflare AI Gateway |
| Mistral | Transcribes and diarises audio (Voxtral models) | The audio of each recording while it is being transcribed |
| Provides sign-in and the Calendar and Meet APIs | Sign-in and authorised API requests; we receive the identity, calendar and Meet data described above | |
| Fly.io | Runs voiceprint inference on infrastructure we control in Amsterdam, in the European Union | Audio segments identified only by transcript speaker ids; never names or email addresses |
| Yoco (Yoco Technologies (Pty) Ltd, South Africa) | Takes card and other payments | Your payment details and purchase amount. We never see your full card number. |
Agents, API clients and notification destinations you authorise can also receive data within the access you grant, including Google-imported transcripts and meeting information. They are chosen by you rather than appointed as our infrastructure operators. Before connecting one, check what it receives, where it sends data, how long it keeps copies, and whether its provider complies with Google's Limited Use restrictions. Permission to access your account is not permission to use Google data for advertising, sale or general model training. We retain our own obligations for any Google data we transfer.
We will also disclose information if a court or a competent authority lawfully requires it. Where we are allowed to, we will tell you first.
6. Where your information goes
Some of our operators store or process information outside South Africa. Section 72 of POPIA lets us do this where the recipient is bound by law or contract to protect your information to a standard like POPIA's.
- Mistral processes audio on servers in the European Union, where data protection law is recognised as adequate.
- Cloudflare, OpenAI and Google may process information in several regions. Fly.io processes voiceprint inference in Amsterdam, in the European Union.
- For an agent or integration you authorise, processing locations depend on the recipient and any AI provider it uses. Check those arrangements before sharing data.
7. How long we keep it
- Recordings, transcripts, rosters and voiceprints: until you delete them or close your account. Deleted items leave the live service immediately and are purged from backups within 30 days.
- Imported Google Meet transcripts: kept as copies in Tymbr until you delete them or ask us to delete them. Google's limited API availability does not make those copies expire. Calendar events and imported participant details are kept to support your meeting history; email hello@tymbr.dev to request deletion of imported data that has no delete control.
- Device credentials: until you revoke them or close your account.
- Purchase and usage records: five years after the transaction, to meet tax and accounting law.
- Technical logs: 30 days, unless we need a specific entry to investigate a security incident.
To stop future Google imports, remove Tymbr's access from your Google Account's third-party connections. Revoking access does not delete previously imported data or remembered voices. Delete those separately in Tymbr, or contact us for help. We do not automatically mirror changes or deletions in Google, and deleting a Tymbr copy does not delete Google's original. Revoking an agent or removing a notification destination also does not erase copies already held by that recipient; request deletion from it separately.
AI providers may retain data separately from Tymbr. OpenAI's API data is excluded from model training by default unless data sharing is enabled, and its standard abuse-monitoring retention can be up to 30 days, subject to its published exceptions. Disabling our gateway logs does not establish Zero Data Retention at OpenAI. See OpenAI's API data controls and Cloudflare's Workers AI data policy.
8. How we protect it
Every recording, transcript, roster and voiceprint is tied to one account, and every read and write in our systems checks that account first. Data is encrypted in transit and at rest. Voiceprints are additionally encrypted before storage. Access to production systems is limited to the Information Officer and protected by multi-factor authentication. If we become aware that your information has been accessed without authority, we will notify you and the Information Regulator as section 22 of POPIA requires.
9. Your rights
Under POPIA you may, free of charge unless the law allows a fee:
- ask whether we hold personal information about you and get a copy of it;
- ask us to correct or delete information that is inaccurate, out of date or no longer needed;
- object to processing based on our legitimate interests;
- withdraw consent to voiceprint processing at any time, without affecting anything done before;
- complain to the Information Regulator if you are not satisfied with our response.
Most of these you can do yourself in the web app: download, edit and delete recordings, transcripts, rosters and voiceprints, and close your account. For anything else, email hello@tymbr.dev. We will reply within 30 days. If you appear in someone else's recording, the account holder is the responsible party for it. Contact them first. If you cannot reach them, contact us and we will forward your request to them within five business days, and act ourselves only where the law requires us to.
10. Cookies
We set only the cookies needed to keep you signed in. There are no analytics or advertising cookies. Clearing cookies signs you out and nothing more.
11. Changes
We will update this policy as the service changes, for example when we add a payment processor. Material changes will be emailed to you at least 14 days before they take effect, and the date at the top will always show the current version.
A policy update does not by itself authorise a new use of Google data. We will obtain separate consent before beginning a use that needs it.
12. Contact and complaints
Information Officer: Gareth Saul, trading as 20 East
Email: hello@tymbr.dev
Address: 18 Lyndhurst Road, Kenilworth, Cape Town, 7708
Telephone: no telephone line; email hello@tymbr.dev for questions and complaints
If we do not resolve your concern, you may complain to the Information Regulator of South Africa:
The Information Regulator
Woodmead North Office Park, 54 Maxwell Drive, Woodmead, Johannesburg, 2191
Email: POPIAComplaints@inforegulator.org.za
General enquiries: enquiries@inforegulator.org.za