HivemindOS manual
HivemindOS Cloud Agent Calls
Cloud Agent Calls are the paid, room-based version of agent voice. The free version already covers the common case: one person talking to one agent.
Free versus paid
The free call path is BYOK: one user, one selected agent, the user’s own OpenAI Realtime key, and a direct call from the dashboard or phone. See Calling.
A call with several people or several agents needs a real room, and a room needs managed infrastructure. HivemindOS creates the room, issues access tokens for each participant, brings the call agent in, keeps it running, and routes audio between everyone. That is an operated service, so it is paid.
| Capability | BYOK Agent Calls | HivemindOS Cloud Agent Calls |
|---|---|---|
| One user calls one selected agent | Included | Included |
| Uses user’s OpenAI Realtime key directly | Yes | Optional |
| Managed LiveKit rooms | No | Yes |
| SFU audio routing | No | Yes |
| Mobile-friendly room tokens | No | Yes |
| Multiple humans in one call | No | Yes |
| Multiple agents in one call | No | Yes |
| Hosted room reliability | No | Yes |
| Premium paid service | No | Yes |
The call agent can still hand work to the agent on your computer, so a room call is still an agent call, not a separate voice assistant.
Why rooms need a server
Rooms use an SFU (selective forwarding unit): a switchboard server. Each participant sends audio once to the server, and the server forwards it to everyone else. Without one, every participant must connect directly to every other participant, which breaks down quickly. With one, a room can hold a user on phone and desktop, a friend, and several agents at once.
Product promise
The promise is not “pay to talk to your agent.” It is “pay when HivemindOS runs the room for you.” Free BYOK calls stay the default. The room provider and voice-model credentials belong to the managed HivemindOS Cloud service, not to your local setup.