Every conversation with an agent can be spoken instead of typed. Tap the mic in the chat, say what you want, and your words land in the conversation as your message. The agent answers in writing and out loud.
Say it
Open any chat, from an agent’s Chat button on the Team page or from Open Collaboration in the top bar. Next to the attach button there is a mic.
- Tap it and start talking. The first time, one line tells you what happens to your voice, then your browser asks for permission to use the microphone.
- Tap it again when you are done, or just stop talking. After a short silence the recording ends on its own. One message can be up to a minute long.
- Your words appear as your message, the same as if you had typed them, and the agent gets to work.
If the agent heard you wrong, say it again. The transcript is on screen, so you always see what it heard before anything happens.
Hear the answer
When you spoke your message, the reply is read aloud in the agent’s own voice as soon as it arrives. Every agent has a voice of its own, so your Producer does not sound like your Publicist, and in a conversation with several people you can tell who is speaking.
Every reply also has a Listen control under it, whether you spoke or typed. Playing a reply pauses whatever else is playing in SongPilot, and the song player pauses a reply the same way.
In a group chat, the person who spoke hears the reply. Everyone else reads it and can play it, and playing it again costs nothing.
Saying yes to a price
Agents still ask before spending credits. When one quotes a price, it says it out loud with the amount in words and asks whether to go ahead. Say yes and it runs; say anything else and it does not. The yes only counts in the message right after the quote, the same as when you type, and you can always tap the yes instead.
What is kept
Only the words. Your audio is turned into text by our speech model and is not stored. The transcript is kept as your message, like any typed message, and deleting a conversation deletes it the same way. The agent’s spoken reply is kept for a day so it can be replayed, then it is gone.
Messages that were spoken carry a small mic mark, so in a group everyone knows those words were heard rather than typed.
What it costs
- Listening: 1 credit for every 10 seconds you speak, rounded up.
- Speaking: 2 credits for every 100 characters an agent says, rounded up, charged once per reply. Replays are free.
- The agent’s work itself is charged as a normal chat turn.
If your workspace runs on its own key for image and video generation, listening and speaking cost no credits.
Give a custom agent a voice
Agents SongPilot provides have fixed voices that match who they are. An agent you created speaks with one of those same eight voices: it starts with the voice of its role, and you can change it on the Team page in the agent’s settings. None of the voices is a real person’s, and there is no way to upload or copy one.
Browsers and phones
Voice works in current Chrome, Safari, Firefox and Edge, on desktop and on phones. On a phone, keep the SongPilot tab in front while you are speaking; switching apps ends the recording. If your phone says the microphone is unavailable, reload the page.
