Skip to content

Recordings and retention

Calls are recorded in two channels: the person on one, and what the person hears on the other. The recording is a WAV file. To switch recording on or off for a campaign, see recording.

How a recording is kept safe:

  1. The call is written to a temporary spool on the phone server.
  2. It is uploaded to storage. The storage service reports a SHA-256 fingerprint and a size, and both are checked.
  3. Only then is the spool file deleted.
  4. If an upload fails, it is retried. If the phone server stops mid-call, the file it left is repaired and registered when it starts again.
  5. A reconciliation check lists answered calls with no stored recording after 15 minutes, a failed recording, or one still waiting. Platform operators read it.

Owner, admin, supervisor, QA and compliance can play or download a recording. Every access is written to the audit trail first. If the audit entry cannot be written, nothing is served. Recordings are found from the person’s journey. See journeys.

In some states everyone on the call must agree to recording. The rule recording_consent with all_party marks them. When any state a person points to needs it, the call carries that flag, and the bot asks for recording consent. The Safety overview’s check says “Everyone on the call must agree to recording here.” Calls placed by hand through the API, and every inbound call, get the strict setting.

All-party states in the starting rules: CA, CT, DE, FL, IL, MD, MA, MI, MT, NV, NH, OR, PA and WA. Counsel should confirm the list and the wording. See rules and counsel review.

Retention is how long you keep recordings before deleting them. Counsel should decide it, and it can differ by what the recording is evidence of.

Until these are built, decide your retention period with counsel, write it down, and plan to delete by hand. See before you call.