Live voice,
three calls in.
Your server opens a cast and mints a token. Your users talk. Nothing is recorded, nobody needs a Soapbox account, and test keys are free.
// 1. your server opens a cast const cast = await soapbox.casts.create({ title: 'Thursday standup', }); // 2. mint a token for one of your users const { token, url } = await soapbox.tokens.mint(cast.id, { identity: user.id, displayName: user.name, }); // 3. the browser joins with it const session = await join({ token, url }); await session.start();
One moment
Three calls
Open a cast, mint a token, join. No rooms to provision and no media servers to run.
Nothing recorded
Audio is relayed and never stored. Chat lasts three minutes. There is nothing to leak.
Built for agents
OpenAPI, llms-full.txt and an MCP server, so a coding agent can wire it up against a test key.
The limits are the product
Designed to end. Built to be forgotten.
4h
The longest a cast runs, then it ends on its own.
50
People in each cast. Eight on stage.
10
Casts at a time, per account.
0
Recordings, transcripts or replays. Ever.
Warnings come at 30, 10, 5 and 1 minute, so a cast is landed rather than cut off mid sentence. If a feature you are planning needs an archive, it cannot be built here. Better to know that now than three days into a build.
What you are building on
Voice, at the same time, anywhere you put it.
A cast is a live room with people talking in it. Your users join from a browser, a phone, a game, a kiosk, whatever you are shipping, and they never need a Soapbox account. What they say reaches each other and nowhere else.
Test keys are free
They create real casts and are never billed, so working it out costs nothing.
Written for agents
An OpenAPI contract, an llms-full.txt, an MCP server, and build prompts you can hand straight to a coding agent.
A browser SDK that is not a microphone
Subscribe only, so an embedded cast has no permission prompt and cannot put somebody on a stage they did not choose.
Roles, not plumbing
Two hosts, eight guest seats, and a hot seat that hands somebody five minutes and takes it back. Yours to use or ignore.
And what it refuses
Say these out loud before you build, because none of them are coming:
- No recording. Audio is relayed and never written down.
- No transcript, no replay, no archive of a finished cast.
- No chat history. A message lasts 180 seconds for everybody.
- No video and no screen sharing.
That is the product, not a roadmap. It is why there is nothing to leak, nothing to subpoena and nothing to moderate after the fact, and it is why the whole thing is three calls instead of thirty.
The reference client
We built one. Open a cast and say it.
Free, in a browser, nothing to install. It runs on the public API with no private endpoints and no shortcuts, so what it does is a demonstration rather than a claim.
None of those are features. A cast is a room with voices in it and you decide what it is for. Tag one with up to five words of your own, which is how it gets filtered and found. There is no list of topics to pick from, because the only thing that knows what a cast is about is whatever opened it.
It is a browser tab
Nothing to install, nothing to approve, no store to wait on.
The panel at the top of this page is the app, live, with whatever is on air in it right now. Open it and you are in the same client every example here is written against.
Open SoapboxThe hot seat
Five minutes on the microphone. At five minutes Soapbox puts you back in the crowd, finished or not.
Chat that forgets
Every message lasts 180 seconds. Nobody screenshots a thread that is already gone.
A raised hand is a signal
Not a queue. Hosts pick whoever they want, in whatever order they want.
Nothing to install
It is a browser tab. The same tab on a phone, where it keeps playing as long as the screen is on.
If signup says we are full
Soapbox holds a thousand accounts on the reference client while it is young, so you are almost certainly straight in. If it is full, leave your address and you get one email when a place opens. No newsletter, no drip campaign. The cap is on this app and not on the API: a key is a different door.
Pricing
Pay for the minutes people spend.
A participant-minute is one person in a cast for one minute. That is how the audio is billed to us, so it is how it is billed to you.
Work it out
That comes to
$10.89
a month, on live keys
- Participant-minutes
- 10,800
- 1,000 for $0.99
- 11
- Minutes bought
- 11,000
- On test keys
- $0
Test keys
Free
never billed
Real casts, every endpoint, no card and no expiry.
Production
$0.99
per 1,000 participant-minutes
5 free minutes to start. Buy as many packs as you need.
The reference client
Free
in your browser
Host or listen, every cast feature, on the same API.
20,000 minutes
$14.99
$0.75 per 1,000, a 24% saving
50,000 minutes
$34.99
$0.70 per 1,000, a 29% saving
The bundles add to a purchase rather than replacing the packs, so 21,000 minutes is one bundle plus one pack. The balance refills to the full amount every period and unused minutes do not roll over. If you run out part way through, a top-off is the same prices charged once.
Common questions
What people ask first.
What is Soapbox?OpenClose
A voice API. Your server opens a cast and mints a token for one of your users, they join from a browser or anything else that speaks WebRTC, and they talk. A cast is over within four hours and nothing is recorded.
How do I start?OpenClose
Sign in at /developers and a test key is waiting, with no card and no expiry. Test keys create real casts and are never billed, so working out whether this fits costs nothing.
Do my users need a Soapbox account?OpenClose
No. You mint a token for whatever you already call that person, and the token is all their browser ever sees. They never meet Soapbox and never see our name unless you choose to show it.
What happens when I hit ten casts?OpenClose
The API refuses the next one with a clear error, so you can tell your user. The reference client puts people in a line instead. That is the only difference between the two.
Can I use it from React Native?OpenClose
Yes, with livekit-client directly. The browser SDK attaches audio to the document, which React Native does not have, so the SDK is the one part of this that is browser only.
Is anything recorded?OpenClose
No. Audio is relayed live and never written to storage, so there is no archive, no playback and no transcript. When a cast ends there is nothing to go back to, for you, for us, or for anybody using the API. That is the product rather than a feature we have not built.
Why four hours?OpenClose
A cast is a moment rather than a channel, so it ends on its own. Warnings come at 30, 10, 5 and 1 minute, which is enough to land it rather than be cut off mid sentence.
What happens to chat?OpenClose
Each message lasts 180 seconds for everybody, then it is gone. It is never written to a disk, so there is no thread to screenshot and nothing to go back and read.
What is the app at /feed then?OpenClose
The reference client. It is free, it runs on the same public API with no private endpoints, and it is there so you can see what the calls add up to before you write any. Ten casts exist at a time on it.
Does it work on iPhone?OpenClose
In the foreground. Safari stops the sound when the screen locks, so keep the cast on screen. Android keeps playing in the background.
Test keys are free, forever.
Production is $0.99 per 1,000 participant-minutes. Buy packs as you grow.