Someone has read the advice. They know to name the problem early, ask for what they need and stay steady when the other person pushes back.
Then the moment comes, and the words are harder to find.
A practice room gives people somewhere to try the words before the real exchange. Now you can offer that room inside your own product.
The Conversations API, our first public API, went live on October 25. It gives your users a room to practice in, a cast that answers back, a Coach in every turn and a saved record of what happened. You build the experience around it. We run the room.
What your users experience
A learner picks a situation. A salary conversation. Feedback for a colleague who missed a deadline. A project meeting where two people want different things. They step into the room and respond in their own words.
Choose from 52 AI characters, each with their own personality, and put up to five of them in a room. They answer in character. Characters respond to the whole conversation, and their reactions shift as it unfolds. Each character’s mood carries from turn to turn. With each turn, the API can report each character’s expression, one of 16 named states or a simpler set of five, so your app can show faces that change as the conversation moves. Using them is optional. It also reports whether each character is present, has stepped away or has left the room, and a character pushed too far can leave for the rest of the session. In ready-made rooms, one person may arrive having a particular kind of day, a phone call or time check can interrupt, and someone may step out and come back. After the session, the debrief shows how each character’s responses changed. The Coach works inside every turn, helping the learner think about the next move while the moment is still fresh. The learner can press on, try a different approach or let the room push harder.
When the practice closes, it is saved. The learner can replay the conversation, read the debrief and come back to observations across sessions, each tied to the moments that mattered.
Choose. Practice. Reflect. Try again.
Watch the two-minute API film and read its transcript.
Why it is powerful
The whole room. Many of the conversations people dread involve more than one other person. A room can hold one participant or five. Each responds in character, and every saved turn records who said what, so your interface can show the conversation the way it happened.
A Coach in every turn. Guidance arrives inside the conversation, not after it is over. You choose the coaching lens, and the Coach’s work is included in the price of the turn.
Controls that shape the practice. Set the challenge and the emotional heat. Start supportive, then raise the stakes as the learner finds their footing. The controls can change from one turn to the next.
Ten languages. The cast speaks ten languages, so a learner can practice in the language they live or work in. The Coach stays in English.
Voices when you want them. Add spoken responses to any saved turn: each participant’s voice and the Coach’s spoken guidance, in one generation.
Practice that carries forward. Replay, debrief and observations across sessions are part of the API. Session continuity and saved preferences let a learner pick up where they left off.
One foundation. The API runs on the same patent-pending foundation as our Conversations app.
How your team integrates it
Three steps.
1. Prepare a room. Read the catalog. Choose one of 100 authored rooms in eight categories, or set up open practice on a topic of your own with a supported cast. Pick the participants and the controls. Your server creates the session, and setting up an empty room is free.
2. Open it and send turns. Ask the room for its opening, then send each of the learner’s turns. The room’s response streams back to your application. Generation requests use an idempotency key, so your server can recover the original request after a dropped connection. The guide explains when to check its status and when to retry.
3. Read the saved practice. Fetch the playback, the debrief and the insights across sessions whenever your product needs them. Reads carry no generation charge.
Your application owns the screens, the user journey and its relationship with its users. Your server holds the API key and uses an opaque identifier for each user. We run the cast, the Coach and the saved practice.
The integration guide walks through each step with TypeScript, Python and cURL examples, a downloadable OpenAPI contract, audio, export and deletion.
One simple price
A billable text turn costs $0.03. That covers every participant in the room and the Coach. Optional spoken responses add $0.05, for $0.08 a turn. There is no per-speaker fee: five people in the room cost the same as one.
Ten text turns cost $0.30. With spoken responses on each, $0.80.
Usage is paid from prepaid credit in your developer workspace, which shows your rate and allowance. Past usage is never repriced, and self-service price increases come with 30 days’ notice. The pricing page has the details.
Who can use it
Self-service access is open to adults in the United States and Canada, excluding Quebec. Institutional, children’s and companion uses require an agreement, so tell us what you have in mind. The API is not offered for health uses.
Where it fits
If you want to practice a conversation yourself, Conversations is our app for that. The Conversations API is for builders bringing practice to their own users.
Start building
Look at the Conversations API, follow the integration guide and check the pricing. When you are ready, add the API to your workspace and send your first turn.
Then ask the question that matters most for your product: where would your users like to try the words before they need them?