Calling

Calling lets you speak with an agent instead of typing. You can start a voice conversation in the dashboard, configure an agent’s voice, and pair supported phone calling when you want an agent to reach you away from the computer.

Voice does not create a separate agent. The call uses the selected agent’s identity, permissions, model, and available context.

Start An In-App Call

1. Open the agent in Fleet or Chat.
2. Choose the call action.
3. Review the microphone and voice provider shown in the setup.
4. Allow microphone access when your operating system or browser asks.
5. End the call when you are finished; review any saved result or follow-up task before acting on it.

The desktop app also offers a hands-free voice channel to Queen Bee. It shows live transcription while you speak and while Queen Bee replies, so you can catch misunderstandings without relying only on audio.

Choose How Voice Is Provided

HivemindOS can use different voice paths depending on what you have configured:

• Bring your own provider uses your approved realtime, speech-to-text, or text-to-speech setup.
• On-device speech can provide a simpler speaker fallback when a supported local voice is available.
• HivemindOS Cloud Agent Calls is an optional managed room path for eligible accounts and is not enabled by default.

The setup surface shows what each path supports before you choose it. A local provider can keep more of the audio path on your machine; a cloud realtime provider sends audio to that provider under its terms. Review the provider choice for sensitive conversations.

Configure An Agent’s Voice

Open the agent’s settings and choose Calls. From there you can:

• select an available voice;
• test the microphone and speaker path;
• choose supported speech-to-text and text-to-speech providers;
• pair or remove a supported phone connection;
• set the spoken prompt used for a scheduled or triggered call;
• review whether the required provider setup is complete.

If a credential is missing, the calls panel opens the relevant setup in place. Shared voice-provider setup can be stored through Shared Env without putting the plaintext value in Hive Superbrain.

Phone Calling

Phone calling requires an explicit pairing or managed call setup. Pairing links the intended device; it does not grant every agent permission to ring it.

Use the Phone view to review paired-device state and saved call prompts. An agent or automation can prepare a call only through the available calling path and its approval policy. Keep emergency, financial, medical, and other high-stakes decisions out of unattended voice automations unless the workflow has an appropriate human review step.

An in-dashboard call and a phone ring are separate actions. Starting one should not silently trigger the other.

Voice Runs And Recipes

A voice run is the record of one call workflow: what started it, which agent handled it, which tools were available, whether the call completed, and what follow-up was produced.

A recipe is a reusable call pattern, such as:

• a daily spoken briefing;
• an approval reminder;
• a project status check-in;
• a hands-free capture prompt;
• a company update that creates a Work task for anything unresolved.

Review a recipe before enabling it on a schedule. A spoken request can still require confirmation before it sends a message, spends money, publishes, or changes an outside account.

Privacy And Safety

• The call uses the selected provider; review whether audio leaves your machine.
• Live transcription can be wrong. Confirm names, amounts, addresses, dates, and destructive instructions before acting.
• A paired phone is an approved destination, not proof that the current speaker has authority for every action.
• Provider keys belong in setup fields or Shared Env, never in a spoken prompt or transcript.
• Ending a call stops the conversation; it does not automatically cancel a separate task already accepted by an agent.
• Cloud calling stays off until you explicitly enable an eligible managed path.

If calling fails, first check microphone permission, the selected agent’s provider setup, and the paired-device state. Then open Troubleshooting or use text Chat while the voice path is unavailable.