Privacy statement
Look Who's Talking is pre-beta. It is an unfinished product under active development. Features and these terms may change, and this page will be updated when they do.
This statement describes what the Look Who's Talking app and its processing server actually do with your information. It is written to describe what actually happens — in our code and at the third parties we use — not to sound reassuring.
What the app collects
- Audio you choose to record or upload. The app records audio through your device microphone, or accepts an audio file you pick. That audio is uploaded to the processing server.
- The results the server produces: a transcript split by speaker, a speaker timeline, and speaking-time statistics.
- The name or label you give a recording, which is used as the recording's filename.
- Your account details. Using the app requires an account. The app collects the email address you sign up with and an internal user identifier (a UUID) that names your account. If you sign in with Google, Google also returns your name and profile picture, and those are stored with the account. The app itself never stores your password; if you use email sign-in, the hashed password is held by Supabase.
Sign-in is handled by Supabase, a third-party authentication and database provider. If you connect a cloud storage account, that sign-in is separate and happens directly between your device and the provider; the access token is stored on your device, not sent to the processing server.
Your account, and where it is stored
Account records — your email address, the internal user identifier, and,
when you sign in with Google, your name and profile picture — are stored in
Supabase's hosted Postgres database at
howvfsrwwijoflflpziq.supabase.co. Supabase also holds the hashed
form of any email password. Unlike the job results described below, account
records are not ephemeral: they persist until the account is
deleted. The processing server verifies a short-lived signed token issued by
Supabase on each request, so it receives the user identifier but not your
password.
Billing is not enabled in this version: no payment details are collected and no subscription is active.
What happens to your audio
Your audio is uploaded to the processing server at
look-whos-talking-production.up.railway.app, which is hosted by
Railway, a third-party hosting provider. Our server is not the only place your
recording goes. Speaker diarisation — working out who spoke when
— and transcription — writing down what was said — are both
performed by a third-party speech-processing service on servers
located in the European Economic Area (EEA): our server sends
the audio to that service, which processes it and holds it in temporary storage.
That service's core services and storage — its API servers, databases and storage — are hosted on infrastructure provided by a subprocessor in the EEA. Its default is that the processing of a job "may run in any region, including outside the EEA", but our account's processing region is set to the EU (European Economic Area), so the audio processing for our jobs is performed within the EEA only.
Our processing providers state that they do not use customer audio to train their models. Their documentation says audio uploaded to them is deleted from their servers within 48 hours. (The software library we use carries a warning giving a shorter period of 24 hours; where the two figures differ, this page states the longer one.) Separately from the audio, the results they produce — the "who spoke when" segments and the transcript text — are kept on their side for 24 hours after the job finishes, and then deleted.
As soon as processing finishes — whether it succeeded or failed — our server deletes the uploaded audio file. Our server does not use the audio to train anything and does not keep it as an archive. One caveat, stated plainly: if the server process is killed mid-job (for example, a redeployment interrupts an upload), an audio file can remain on the server's disk until the container is replaced.
What is kept on the server, and for how long
The server keeps the results of each job: the transcript text, the speaker timeline, the speaking-time statistics, and the recording's filename. It also keeps a record of each job's status. These are held as files inside the running server container.
That storage is ephemeral. It survives only for the lifetime of the current deployment, and every redeployment of the server destroys it. There is no automatic deletion on a timer within that lifetime, so a result may sit on the server for as long as the deployment keeps running. The server is not the store of record for your conversations — your device, and your own cloud backup if you enable one, are.
The server also writes ordinary operational logs while it works. These record the job identifier, the filename supplied, and processing progress and counts. They do not contain your audio or your transcript text.
Account records are the exception to the ephemerality described above: they live in Supabase's hosted database, not inside the processing container, and persist until the account is deleted. See Your account above.
Where your results and backups live
Results are kept on your device, in the app's local storage, until you delete
the conversation. The audio file is also kept on your device so it can be played
back or shared; after 90 days the app offers to delete the audio of older
conversations, leaving the transcripts and insights in place. Cloud backup is
opt-in and off by default. If you connect Google Drive, Dropbox
or OneDrive and turn syncing on, the app copies your conversations into an
app-specific folder in your own cloud account (for example,
/Apps/Look Who's Talking/ in Dropbox). The developer does not own or
control that storage and does not receive a copy of your backups.
What is shared, and with whom
- Insights and Q&A: if you use the app's "insights" or
"ask a question" features, the transcript text — the spoken
words with speaker labels, not the audio — is sent by the server to
DeepSeek (
api.deepseek.com) to generate the summary or answer. These features only send anything when you use them. - Hosting: the audio and results are processed on Railway's infrastructure as described above.
- Diarisation and transcription: the audio is sent by our server to a third-party speech-processing service located in the European Economic Area (EEA), whose core services and storage are hosted there and whose processing for our jobs is configured to run within the EEA only, to work out who spoke when and to write down what was said, as described above. Our processing providers state that they do not use customer audio to train their models.
- Analytics (PostHog, as of this version): the app sends
product-usage events to PostHog Cloud, which is hosted in the EU and is
reached at
eu.i.posthog.com. Those events record which screens you open and which features you use, together with an identifier for the app installation (an anonymous ID, or your account ID if you are signed in). PostHog is never sent your audio, your transcript text, your speaker names, or any other content you record or type, and it receives no personally identifiable information. The app contains no advertising or crash-reporting service. If this changes, this statement will be updated and, where required, you will be asked first.
Your audio and results are not sold, and are not provided to anyone else except as described here.
Your choices, and deletion
You can delete a conversation, and its locally stored audio, from within the app. Deleting a backup is done in your own cloud account.
You can delete your account at any time. In the app, open
Settings, tap Delete my account, and type
DELETE to confirm. That erases the account record, the usage
history held with it, and the conversations stored on that device. It is
permanent and cannot be undone.
If you cannot open the app, you can ask for your account to be deleted by emailing privacy@look-whos-talking.ai from the address you signed up with, or by following the instructions on the delete-account page, which restates both routes. Deleting the account removes your account record and any usage history held with it. Conversations you backed up to your own cloud storage are not covered by an account deletion — delete those in your cloud account.
Changes
This is a pre-beta product. This statement may change as the app changes, and the "last updated" date below will be revised when it does.