Whether a call is recorded
Six settings can switch recording on, and the call is recorded if any of them says so. They are not a hierarchy and none overrides another:
The call’s own page names which level did it, so “why was this recorded”
has an answer without opening six screens. On a call that was not recorded,
the same line reads no level of this call has recording enabled.
The one way it can fail quietly
A node needs somewhere to put the file. If recording is on and the node has no recordings directory configured, nothing is recorded and the panel looks exactly as it does when recording is off. The daemon says so at warning level and the call’s page carries enabled, but this node has no recordings directory. An operator fixes it by settingFIRETONE_RECORDINGS_DIR on that
node.
Where the audio lives
A recording starts as a file on the node that took the call, named after the call. Where it goes next depends on whether the organisation has given FireTone a bucket of its own:- With cloud storage configured, the file is handed to the organisation’s own bucket. It is then reachable from any node, survives a node being rebuilt, and sits on storage the organisation controls — which also means they can apply their own lifecycle rules to it if they would rather answer the retention question themselves.
- Without it, the file stays on the node that took the call. That works, and it has three limits worth knowing: one disk, unreachable from other nodes, and gone if that node is rebuilt.
recordings:listen.
How long anything is kept
Three windows, set per organisation, because these are three different exposures with different answers. A recorded sales call and a member of the public’s voicemail are not the same risk.Nothing is deleted unless somebody asked. Recording and voicemail
windows are empty until a number is typed, and empty means keep. This is
deliberate: software that starts destroying a tenant’s recordings because
an upgrade invented a default is worse than software that keeps too much.
The transcript window is the exception — it has carried 365 days since the
store was created, and that is a default, not a recommendation. A tenant
with a shorter obligation should lower it.
What deletion actually does
A sweep runs every hour and takes a bite at a time rather than one long delete, so a busy system is not locked while it runs. The file and the row go together. A recording of somebody’s voice that nothing points at any more is the worst kind of file to leave on a disk, and a row pointing at a file that has gone is a broken player in the panel. So each is removed with the other — and where audio has moved to an organisation’s bucket, the row is cleared only once the object is gone. If the bucket cannot be reached, the row stays, on purpose: an orphaned recording is exactly what this is meant to prevent. Deleting a recording does not delete the call. The call record, its duration, its cost and its outcome stay; what goes is the audio and the transcript.Who can hear what
Reading and listening are different permissions, and they are different disclosures:
A person with
conversations:read and no recordings:listen can read a
transcript and cannot hear the voice. That is the intended shape: the text
answers “what was agreed”, the audio is the recording of a person.