# Live voice (Hermes Desktop)

**What this is:** talking to Bill or Ben by voice outside a meeting, through the Hermes Desktop app.
**Who it is for:** staff who want a spoken conversation, and admins who set up the voice mode.
**You will learn:** the two voice modes, what they cost in settings, what an agent can and cannot do in a voice session, and what is not wired yet.

## Use it

1. Install Hermes Desktop and sign in to the agent's dashboard with the login your Bill and Ben contact gives you. The app connects only over a private connection (for example a VPN) that your Bill and Ben contact provides. This connection is not verified yet.
2. Speak. You can interrupt the agent while it talks.
3. Do not expect tools. See [What a voice session can do](#what-a-voice-session-can-do).

### Two voice modes

| | Chained (default) | Realtime |
|---|---|---|
| Speech to text | a speech model chosen by the platform | the realtime vendor |
| Thinking | the agent, one turn at a time | the vendor talks and hands work to the agent |
| Speech out | a voice model chosen by the platform | the realtime vendor |
| While work runs | silent | keeps talking and says briefly what it is doing |
| Needs | nothing extra | an account with the realtime vendor |

Your Bill and Ben contact picks the mode and the voices for Bill and Ben.

### What a voice session can do

Agents can only use their tools when Bill and Ben have checked that you are allowed to ask. A voice session in the desktop app does not get that check, so the agent can **talk** but its tools are blocked, except Helmora tools, which check access to their own boards. It is not verified whether Hermes Desktop sessions get that check some other way. For work, use [Teams chat](teams-chat.md), a [meeting](meeting.md) or the [Helmora](../canvas.md).

## How it works

In short: your speech becomes text, the agent answers, and the answer is spoken back.

<div class="bb-diagram" role="region" tabindex="0" aria-label="Diagram: how a spoken sentence in Hermes Desktop becomes a spoken answer. Scroll sideways if it is wider than the page." markdown="1">

![how a spoken sentence in Hermes Desktop becomes a spoken answer](../diagrams/public/live-voice.light.svg#only-light){ style="--bb-diagram-w:433px" }
![how a spoken sentence in Hermes Desktop becomes a spoken answer](../diagrams/public/live-voice.dark.svg#only-dark){ style="--bb-diagram-w:433px" }

</div>

[Open full size](../diagrams/public/live-voice.png){ target="_blank" rel="noopener" }


The dashboard is the agent's own web page, shared by everyone who uses the same agent. It does not know who you are, so there is no per-person identity in this path. Meeting speech is a different path ([Meeting](meeting.md)).