Quick answer
A programmable conference is a multi-party voice room your application controls through an API — create the room, dial participants in, mute, hold, lock, add an AI agent, start recording, and close the room, all as API calls while the conversation is live. It is the opposite of a 1:1 call: instead of one caller talking to one callee, the room mixes N participants at once, and the controls on the room — mute, hold, lock, record — are API requests your backend makes, with every change visible to operators on the dashboard. Devotel Orbit ships the full conference surface — 24 conference operations in the public API, a dedicated Voice › Conferences area in the dashboard, live ringing and waiting indicators, per-participant consent, and recording lifecycle controls — so a conference is a building block, not a feature-gated upgrade.
The unit is a room, not a call
The one-to-one call is a line: two endpoints, one media path, one outcome. A conference is a room: a mixing surface that any number of participants join and leave, where the application — not a phone keypad — decides who is in, who is muted, who is recording, and when it ends. That difference changes what voice can do:
- The room outlives any single participant. Participants join and leave while the conference continues; the room closes when your application ends it or when everyone has gone.
- Every action is a control-plane call. Add a participant, mute a noisy line, hold a participant, lock the room against late joiners, whisper to an AI agent, or start recording — each is one API request against the running room.
- The state is observable in real time. The room appears in the dashboard the moment dialing begins, and per-participant status (ringing, waiting, connected, failed) is visible to operators and readable through the same API your application calls.
Once the unit is a room, warm transfer, three-party supervisor joins, dial-in legal consultations, and on-call bridges all stop being phone-system tricks and become combinations of the same few operations.
What Devotel Orbit shipped in Q3 2026
Two recent changes show where the surface is now. First, conference creation became honest about routing: creating a room now checks that outbound routing is ready before the room exists, so a misconfigured voice application fails the create with a clear error instead of minting a room whose every participant leg immediately fails. Second, the Voice hub now treats a conference as live the moment it starts dialing — the Live Calls panel and the Active Calls headline show the room during the ringing and waiting phase instead of staring at "No active calls" until the first participant answers.
Those two changes sit on a base that was already complete: participant-level failure reasons in the call log and timeline, per-leg SIP failure diagnostics, PIN-protected dial-in with tenant-controlled passes, and conference rows that stay consistent between the list, the detail page, and the analytics rollup.
The API surface: 24 conference operations
The public API carries 24 conference operations under the voice surface, covering the full room lifecycle:
- Create, list, inspect, update, and end rooms, plus aggregate and failure-analytics views over them.
- Participant controls — add and remove participants, per-participant mute/unmute, hold/unhold, and room-wide mute all/unmute all.
- Room access — lock/unlock the room against new joiners, and regenerate the dial-in PIN.
- AI participation — add an AI agent to the room, list and remove them, hold the agent in whisper mode (it hears the room and is not heard by it until you choose), and hand the agent over when it should actually speak.
- Recording and consent — record per-participant consent, run consent collection at the room level, then start, stop, and retrieve the recording.
Every mutating operation accepts a Stripe-style idempotency key, so a retry on a timeout replays the original response instead of double-dialing a participant.
The dashboard surface
The same capabilities exist for operators who never touch the API. The dashboard's Voice › Conferences area lists active and past rooms with searchable participants and PIN status, and the conference detail page shows the room timeline — each dial attempt, each join and leave, each failure reason — alongside participant chips that carry the real per-leg outcome rather than a generic "ended." Because the September changes landed, a room in its ringing phase is visible there from the first dial, not after the first answer.
Where conference plus AI agent plus recording matters
Three workloads buy exactly this combination, because in each of them the conversation itself is the record.
Legal. A client intake or settlement discussion happens in a room of three: the attorney, the client, and an opposing party or witness. The room is PIN-gated so dial-in is controlled, locked once the parties are in, and recorded after consent collection. The AI agent sits in whisper mode capturing objections or action items; consent receipts and the recording's lifecycle (started, completed) arrive as webhooks the matter-management system can file directly. The tenant owns every control — who is admitted, when recording starts, and how long records are retained.
Financial services. Advisors must document advice conversations. A conference between advisor, client, and a compliance observer starts with consent collection at the room level, records from the first spoken minute, and the AI agent transcribes and tags disclosures against the regulated script. The audit ledger on the room (joins, leaves, mute events, recording state changes) is the supervision trail.
Telehealth. Provider, patient, and an interpreter or family member join a room the scheduling system created when the appointment booked. The room locks when the visit starts; recording requires explicit consent collection, and the AI agent handles intake while the provider is with another patient. Recording retention and consent handling stay tenant-configured — Orbit provides the controls; the healthcare tenant's policies decide how they are set.
In all three, the point is the same: the conference is where the regulated conversation happens, so the room's controls — admission, consent, recording, agent participation — are the compliance surface the tenant operates.
Frequently asked questions
When should I use a conference instead of a 1:1 call?
Use a conference the moment a conversation has more than two parties, or when a party may join or leave mid-call — warm transfers, supervisor joins, interpreters, observers. A 1:1 call that later needs a third party is a failed design; start those flows as rooms.
Can an AI agent participate without the room hearing it?
Yes. Whisper mode puts the agent on the room as a listener that can hear participants without being heard, until your application hands it over to speak. That is the standard pattern for live transcription, disclosure monitoring, and advisory presence.
How do I stop late joiners once the room has started?
Lock the room. Locking closes admission while the conference keeps running for the participants already in; unlocking reopens it. For dial-in conferences, the PIN gates admission even while the room is unlocked, and the PIN can be regenerated without recreating the room.
Who controls recording consent?
The tenant does. Orbit provides the consent-collection flow, per-participant consent records with receipts, and the recording lifecycle endpoints; your policies decide when consent is required, and nothing records until your consent rules pass. Recording, retention, and erasure are tenant-owned settings.
Are retries safe if my backend times out during a participant add?
Yes. Send a stable idempotency key with the request; a retried request replays the original response for 24 hours instead of dialing the participant twice, and concurrent in-flight duplicates are rejected with a 409.