Quick answer: Every recording decision a business makes runs across three stacked questions — does the jurisdiction require one-party or two-party consent, which calls does your policy actually record, and how are participants notified. The platform can only ship the controls; the answers to all three are yours, because you are the one recording. On the Devotel Orbit voice stack those three questions map to three tenant-owned controls: the eligibility mode (which calls auto-record), the consent announcement mode (how participants are told), and the recording policy (the selective rules layer targeting queues, agents, regions, and consent state). This post is the industry-side map of why those exist — the legal layer underneath, modelled the way carrier and acquirer policy now frames it.
What changed: consent moved from a legal footnote to a carrier-policy anchor
The rules themselves are old. Federal US law and roughly forty states run on one-party consent: if you are on the call, you may record it. Eleven states and a handful of jurisdictions elsewhere require all-party ("two-party") consent, and the EU routes recording through the GDPR's lawful-basis and transparency duties, which function as the same all-party requirement. None of that is new. What changed is who now reads your recording posture as part of your standing with them:
- Carriers and acquirers fold it into policy enforcement. The 10DLC era taught messaging buyers that registration vetting, throughputs, and quiet-hours-style windows all became costs and obligations a platform passes to the tenant. Voice is on the same slope: the entities terminating your calls — and the platforms sitting between you and them — increasingly ask how your recording posture matches your traffic, not whether your counsel signed a policy page.
- The enterprise questionnaire asks for architecture, not policy prose. A regulated buyer (finance, healthcare, collections) does not accept "we announce where required." They ask: which queues record, which regions force the audible notice, what stops a
denied-consent call from being captured, and who can open the recordings. That is a configuration question, and it is answered with the platform's controls or your own implementation on top of them. - Recording feeds AI now, so the retention math changed. The call that is transcribed, summed up, and scored is a data asset with a real decay curve. "Record everything, keep it forever" stopped being cheap the day it became model input. A selective posture — record these queues, sample these, keep 90 days — is both the compliance answer and the cost answer.
Plan the configuration in two steps. First, the legal map: jurisdictions divide into one-party regimes, where any participant may record the call, and all-party regimes, where every participant must consent. Second, the carrier layer: acceptable-use terms from the carrier or platform layer their own consent asks on top. The legal split sets the floor; the carrier terms set the posture.
What the tenant owns: a per-jurisdiction consent model, expressed in two controls
You are the party of the recording, so the legal analysis is unavoidable. The platform's job is to make the analysis expressible. The questions resolve to two orthogonal axes, and Orbit deliberately exposes them as two separate settings rather than one vague "recording mode":
Axis 1 — which calls record (the eligibility mode).
all_calls— everything is recorded automatically.outbound_only/inbound_only— direction-filtered recording.manual— no automatic recording; the operator opts in per call.
Axis 2 — how participants are notified (the consent announcement mode).
none— no audible notice (correct only in one-party regimes, and then only when carrier policy is satisfied).announce_all— all parties hear the notice.announce_caller— only the caller hears it (your side stays silent).dual_channel— a distinct, per-channel consent pattern where the recording itself keeps the announcement legible on its own channel.
Every combination is a deliberate posture. The classic bad states are the two mismatches: all_calls with none (records everything silently — a liability in all-party jurisdictions), and manual with announce_all (notifies on calls nobody asked to record — an announcement on a call that is not being recorded just burns customer patience). The modal dialog on Voice → Calls → Recording settings lays the two axes out side by side and warns on the first of them.
The per-jurisdiction mapping is tenant work. The US common shapes:
- One-party (the federal default plus ~40 states): recording is lawful because you are a party to the call. Announcement is optional as a legal matter — which is not the same as wise. Most operators announce anyway, because the announcement converts a tense legal question into a settled process question.
- Two-party states (~11, with California, Pennsylvania, Florida, Illinois, and Maryland among the cited examples): an audible notice or equivalent documented consent is required. Whichever states your agents record into, the configuration should provoke the announcement under a region rule, not rely on agents remembering a policy page.
Cross-border tenants add the GDPR layer: recording is processing, so a lawful basis, transparency, purpose limitation, and retention bound all have to exist before the recording is lawful at all — the announcement is not optional framing there. The tenant policy should therefore be per-queue and per-campaign at minimum, not a single global switch.
Configure the posture in Orbit: eligibility, announcement, selective rules
All three are tenant-owned settings on the organisation's recording configuration — and no platform-side hard gate overrides them; everything except the one federal dialing-window guard is tenant-configurable.
1. Set the eligibility and announcement axes (org settings).
Under the organisation's settings.recording object, the dialog writes:
{
"eligibility_mode": "all_calls",
"consent_announcement_mode": "announce_all",
"consentText": "This call may be recorded for quality and compliance purposes.",
"retentionDays": 90
}eligibility_mode— one ofall_calls,outbound_only,inbound_only,manual.consent_announcement_mode— one ofnone,announce_all,announce_caller,dual_channel. When any announcement mode is set, theconsentTextstring is what the caller hears; without an announcement mode, the field is inert.retentionDays— the storage life of recordings before they age out. Regulated setups pair it with the archive path, not with "keep forever."
2. Layer the selective recording policy on top (admin-only rules).
Direction-level eligibility is a blunt instrument: regulated verticals need predicate-level recording. The GET /api/v1/voice/recording-policy and PUT /api/v1/voice/recording-policy endpoints (admin-scoped) replace the whole-document rule set — a firewall-style priority-ordered list of record/skip rules, each with AND-combined conditions over queue, skill, agent, direction, consent_state, country, region, and campaign. A record rule can also carry samplePercent (deterministic per call, so sampling is stable across retries), and require_two_party_consent — the flag that forces the audible announcement whenever a two-party region match fires.
PUT /api/v1/voice/recording-policy
{
"enabled": true,
"defaultRecordPercent": 10,
"rules": [
{
"id": "collections-full-capture",
"name": "Record every collections-queue call",
"priority": 10,
"conditions": [
{ "signal": "queue", "op": "eq", "value": "collections" }
],
"action": "record",
"samplePercent": 100
},
{
"id": "two-party-states-announce",
"name": "Force the announcement in two-party-consent states",
"priority": 20,
"conditions": [
{ "signal": "region", "op": "in", "value": ["CA", "PA", "FL", "IL", "MD", "MA", "WA", "CT", "NH", "MT", "NV"] }
],
"action": "record",
"requireTwoPartyConsent": true
},
{
"id": "never-capture-denied-consent",
"name": "Never record when the consent ledger says denied",
"priority": 5,
"conditions": [
{ "signal": "consent_state", "op": "eq", "value": "denied" }
],
"action": "skip"
}
]
}Rules evaluate lowest-priority-number-first and stop at the first match — so a denied-consent skip at priority 5 overrides a 100%-capture rule at priority 10 only when it matches. Calls no rule claims fall back to the org-level eligibility_mode decision; with no policy at all, the direction-only axis decides verbatim. Zero Data Retention (zero_data_retention, on the same recording settings) is an absolute veto over everything: when it is on, no rule, direction, or per-call override can capture audio.
3. Wire the announcement into the call itself.
The consent_announcement_mode controls the audible recording notice the gateway plays on the call. announce_caller says only the far side hears it — appropriate when an internal agent absolutely is a party and so is your single consenting side under one-party regimes. dual_channel is the posture for recording-aware architectures where the recording leg and the announce leg have to remain independently verifiable. With the record rule's requireTwoPartyConsent flag set on the two-party region map, the announcement cannot be silently downgraded by the org's general announcement setting — the region rule only ever upgrades.
Frequently asked questions
Which US states require two-party consent for call recording?
The commonly cited all-party states include California, Pennsylvania, Florida, Illinois, Maryland, Massachusetts, Connecticut, New Hampshire, Nevada, Montana, and Washington — with nuance per state statute. Model them as a region rule with requireTwoPartyConsent: true so the announcement fires on any call terminating there, rather than relying on agents to remember.
Does one-party consent mean I never need an announcement?
Legally, in a one-party jurisdiction, a participant may record without telling the other parties. But carrier and platform acceptable-use policy, plus customer-experience norms, usually push to an announcement anyway — and regulated questionnaires ask for it. Treat one-party as permission to make the announcement a choice, not an instruction to omit it.
Who is responsible for call-recording compliance — the platform or the tenant?
The tenant. Orbit ships the three controls — eligibility mode, announcement mode, and the selective recording policy — together with the storage and consent-ledger plumbing. Which calls you record, what you say on the call, how long you keep the audio, and which jurisdictions you operate in are your compliance decisions.
Can I record only some calls instead of everything?
Yes. The recording-policy rules let you record a named queue at 100%, declare a defaultRecordPercent for sampling the rest, skip a consent-denied call unconditionally, and combine record/skip across queue, region, agent, direction, and campaign signals. The eligibility_mode axis is the fallback when no rule matches.
What does the announcement mode actually do on a call?
A consent_announcement_mode of announce_all or announce_caller plays the organisation's consentText as an audible notice when the call records; dual_channel runs a per-channel variant for record-aware leg handling. none is the silent posture, valid only where every regime the traffic touches allows it.
The takeaway
Recording consent reads best when it is modelled as two questions the tenant answers with two axes, plus a regulation-aware rules layer: which calls record, and how the parties find out. Set eligibility_mode and consent_announcement_mode honestly for the regimes you operate in, put the all-party states behind a region rule so the announcement fires regardless of org-level silence, and gate denied-consent calls with a skip so the legal story and the system story are the same one. Everything above is a tenant-owned control on Orbit — the platform supplies the levers; you supply the posture.