Converse

Realtime API reference

Converse Realtime API

AsyncAPI 3.1.0 · API version 1.0.0

Full-duplex voice conversations over one WebSocket. Send a JSON start frame first, then binary microphone audio and JSON controls. Authenticate with api_key in the start frame or an Authorization: Bearer header on the WebSocket handshake.

Need exact payload schemas or validation constraints? Download the machine-readable AsyncAPI contract.

Operations

sendClientMessages

send · Send the start frame, microphone audio, and client controls

start, reset, setVoice, toolResult, toolProgress, toolCancelClient, ambience, playbackStopped, playbackPaused, playbackResumed, playbackPauseFailed, microphoneAudio

receiveServerMessages

receive · Receive lifecycle events, assistant audio, transcripts, and tool calls

ready, turn, asr, utterance, done, interrupted, canceled, playbackPause, playbackResume, ack, toolCall, toolCancelServer, voiceChanged, error, assistantAudio

Messages

Client and server frames

MessageTitleSummary
startstartOpens a conversation. api_key may be omitted when the WebSocket handshake supplied Authorization: Bearer <key>.
resetreset
setVoicesetVoice
toolResulttoolResult
toolProgresstoolProgress
toolCancelClienttoolCancelClient
playbackStoppedplaybackStopped
readyready
turnturn
asrasr
utteranceutterance
donedone
interruptedinterrupted
canceledcanceled
playbackPauseplaybackPause
playbackResumeplaybackResume
ackack
toolCalltoolCall
toolCancelServertoolCancelServer
voiceChangedvoiceChanged
errorerror
microphoneAudiomicrophoneAudio
assistantAudioassistantAudio
ambienceambience
playbackPausedplaybackPaused
playbackResumedplaybackResumed
playbackPauseFailedplaybackPauseFailed

Payload schemas and exact constraints remain available in the machine-readable AsyncAPI contract.