Voice

Press the phone button in the chat composer and start talking. You are speaking to the same agent you type to, with the same memory of your business, the same files and the same tools, and it answers out loud.

Voice works four ways, and they share one setup.

A live call

A call opens the voice orb, which pulses in time with the audio and captions the words underneath as they are spoken. Talk over a reply and it stops, remembers where it was cut off and takes up your new point instead. Say "stop" and it goes quiet at once. When it needs a moment to look something up you hear a hold sound rather than silence.

Live calls are experimental, and Player One says so with a beta notice the first time you press the button.

A voice message

The mic button beside it records a voice message. The recorder takes over the composer while you speak; when you send, a player appears in your own message bubble with the transcript beside it. That transcript is your message, so a short note can steer a reply already running, exactly as typed text does. A very long recording waits for a fresh turn, showing a cancellable row above the composer.

Audio you attach with the paperclip is left alone as an ordinary file. Only recordings you make are transcribed, so an hour of interview is never transcribed by surprise.

Voice messages work from the start. They are transcribed by Player One Provider, which is included with Player One, only transcribes and needs nothing connected. If you would rather another of your accounts did it, pick it on Admin > Voice Models (/admin/voice).

Replies read aloud

Turn on Speak back and answers are spoken as well as written, with the wording underneath to read later. It pauses by itself during a live call, and the call button is unavailable while it is on.

A telephone number

Say "give this agent a phone number", or "ring the courier and ask where the Hanley delivery is".

An agent can take real telephone calls and place them. The number, the telephone network and the speech belong to a telephony provider you hold an account with; Player One supplies the brain and the transcript. Setting it up is a conversation: your assistant asks for the provider's key through a secure prompt, registers a line for the agent, and wires the provider to it. A number you already own with a carrier is imported into the provider rather than moved. Nothing is configured per call: register the line once and every call after that just happens. Your assistant sets a line up and can change it later; Admin > Phone lines (/admin/phone-lines) is where you see it.

Who calls, and where it lands. A call from one of your own numbers joins that agent's main chat and carries your conversation on there, by voice. Anyone else gets a chat of their own, kept for that number, so the second time a customer rings the agent is looking at everything said the first time. A number that is withheld gets a fresh chat each call, because there is nothing to keep it against. A call the agent places for you lands back in the chat you asked from.

You can overrule that per number. Say "put calls from the accountant into my main chat", or name an existing chat for a particular caller, and that is where their calls go from then on. A line can also be set to the older behaviour, a new chat for every call, if you would rather each one stood alone. Routing is not authority: a number pointed at your main chat still has no more say than any other caller. In the chat a call opens with a card naming who called whom and closes with one giving how long it lasted and how it ended. The caller's words appear as their own turns, headed "Caller" with the number, in a panel marked Spoken; what the agent said aloud wears the same mark, and anything it wrote that the caller did not hear sits in a dashed box marked To you, not spoken, so you can always tell exactly what the caller heard. While the call runs, a red banner above the composer shows a timer and a Hang up button, and a pulsing phone badge marks the chat in the sidebar.

You can listen in, and cut in. While a call is live the agent hears the caller, not you, but anything you type into the chat reaches it mid-call, and it answers you in the chat and the caller in speech within the same turn. The agent can also leave you a note beside what it says, "the caller sounds upset", which you see and the caller never hears.

Callers have no authority. A caller is routed by their number, never trusted by it. Whatever a caller says is treated as untrusted, and anything that would need your authority is refused however confidently the caller claims to be you. One live call per chat: a phone call into a chat that already has a browser call running is refused rather than merged.

A call ends when the caller hangs up, when the agent says goodbye and hangs up, when you press Hang up, or after three minutes with no one speaking. The goodbye is the last thing the agent says, and the line can take a few seconds to drop after it. If the provider refuses to end a call from Player One's side, the closing card says so, because the line may still be up.

Where to see the lines and the calls. Admin > Phone lines (/admin/phone-lines), in the Communications group, lists each line with its provider, its agent, its numbers, where callers' words go and whether it is ready or needs registering again. Test runs the checks and shows what each one said, Edit changes a line's settings in place, and Remove revokes it. Below the lines is every call, newest first: when it was, who it was with, how long it lasted, how it ended, a link into the chat it happened in, and Hang up on one still running.

A sandboxed agent can answer the phone too. It gets one thing it can do on a live call, which is to speak, so a public-facing assistant answers in speech rather than typing at a caller who cannot see a screen.

Player One never carries the audio. The provider turns speech into text before each turn and text back into speech after it, so on a phone call the agent only ever reads and writes words. If a line's key might have leaked, ask for the line to be removed and registered again.

The controls

During a call you have mute, the caret beside it for the microphone list, and disconnect. The current microphone is ticked, and choosing another moves the live call onto it without dropping it. Connection status shows in the orb itself. Ask it to speak faster or slower and it keeps the new speed for next time; ask it to hang up and it does.

On the recorder, the cog holds four settings: Microphone, Speak back, Continue listening for a hands-free exchange, and Auto send with the pause it waits for before submitting. The microphone choice applies immediately, even mid-recording, and it is the same choice the in-call caret sets.

Setting it up once

A speech service has to be connected before the first call, and before replies can be read aloud. Ask your assistant to set voice up: it collects the key in a secure prompt, keeps it in the vault and picks a default voice for that agent. Voice messages need none of this, because Player One Provider transcribes them. Your browser asks for microphone permission the first time as well.

Each agent then has its own voice and speaking speed, set on that agent's own page: open Manage > Agents (/agents), choose the agent and use Advanced > Voice. A model the agent chose for itself there wins over the installation's choice on Admin > Voice Models.

Where it lives

The mic, phone and Speak back buttons are in the composer at the bottom of every chat.

Admin > Voice Models (/admin/voice) chooses the installation's voice models, one list per job:

• Spoken replies: the model an agent speaks with, on a call and when it reads a reply aloud.
• Narration: the model for reading something out expressively, such as a voiceover.
• Transcription: files: the model for voice messages and recordings. Player One Provider is listed first, marked Included.
• Transcription: live calls: the model that listens on a call.
• Fast voice: the model a call runs on for an agent with Fast voice switched on. Fast voice stays off for every agent until you turn it on in that agent's voice settings.

On ElevenLabs, Eleven v4 narrates and Eleven v4 Turbo is the quicker of the two for spoken replies. Neither works on a live call, so a call on ElevenLabs uses one that does, Eleven Flash v2.5 for now.

The page's second tab, Health and agents (/admin/voice/health), checks that voice is configured, connected and responding, and lists each agent's voice.

The recorder's own options, and your microphone choice, are saved per browser and stay on the machine you set them on.

Admin > Voice Harness (/admin/voice-harness) is voice on its own, without a chat around it. It stays hidden until you turn on the Advanced toggle on Admin > Runtime (/admin/gateway).

Talk to it puts voice beside the other ways in, Agents covers giving each agent a voice of its own, and Start a voice chat has the steps.