Retention and deletion

What is stored, for how long, how to delete it early, what DELETE actually erases, and who processes your audio.

This page describes what happens to your material after a transcript finishes. None of it is marketing: every line matches a path that exists in the code, and what does not exist yet is stated as not existing.

What is stored

Material	Where	For how long
Uploaded audio (`uploadId`)	Object storage, encrypted	While any non-deleted transcript still uses it
Audio from a `url` of yours	Temporary copy during processing	Removed when the job ends (and within 24 h by the bucket itself, worst case)
Text and words	Database	Until you delete them, or until `retentionDays` expires
Usage record (duration, cost, date)	Database	Permanent: it is what your invoice is built from
Technical access logs	Server log	6 months (Brazilian Marco Civil, art. 15)

Your audio and your transcripts are never used to train models, ours or anyone else's, and never enter an evaluation set.

Setting the period at creation

retentionDays erases that transcript's content on its own, counted from creation:

curl https://api.transcrevo.com/v1/transcripts \
  -H "Authorization: Bearer $TRANSCREVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/call.mp3",
    "speakers": true,
    "retentionDays": 7
  }'

The period is per transcript, not per account, because the material is what decides: a recorded legal meeting and a voice note have no reason to live the same length of time. Accepts 1 to 365 days.

There is no retentionDays: 0. Real zero retention is a processing path that never writes the audio down, and until that path exists here, offering it as a short number would be selling the name without the thing.

Deleting before the period

DELETE /v1/transcripts/{id}

What actually goes:

the text and the words, immediately;
the speaker turns (diarizationSegments), immediately;
the uploaded audio, on the next collector sweep (it runs every 10 minutes and removes the file once no live transcript points at it).

What stays: the usage row, with duration, cost and date. It is the record of a charge that already happened, and erasing it would rewrite your statement for the month. After DELETE, the transcript still answers by id, with text and words null.

A transcript in processing cannot be deleted (409 transcript_processing): the result would arrive afterwards and refill what was just erased. Cancel it first with POST /v1/transcripts/{id}/cancel, which is also never charged.

Backups

The database is backed up, and a deleted transcript can survive in a backup copy until that copy rotates out. No query, no endpoint and no screen reads from a backup: it exists for disaster recovery, it is encrypted, and deleted data leaves it when the copy expires.

Who processes your audio

The sub-processors that touch customer content, and what each one does:

Company	Role
Oracle Cloud Infrastructure	Application and database servers
Cloudflare	Network, attack protection, and encrypted storage of uploaded audio
Vast.ai	GPU machines where audio is decoded and transcribed
Stripe	Payments (never receives audio or transcripts)
Google	Identity verification only, for people who sign in with Google

The full list, with legal bases, international transfers and your rights as a data subject, is in the Privacy Policy.

What we do not offer yet

Zero retention (never writing the audio down at all).
Region selection for processing: the GPU fleet is rented wherever cards are available, and promising a region we cannot enforce would be a promise the code does not keep.
SOC 2 or HIPAA certification. We do not have it, and we will not write that we do.

If one of those is a requirement for your project, write to contato@transcrevo.com: knowing how many people need it is how the work queue gets chosen.