@hamsa-ai/voice-agents-sdk.
Constructor
Code from before public keys passes an API key first:
new HamsaVoiceAgent(API_KEY, options). It still works. See Migrate.
start() options
start() returns a promise that resolves once the call has started. If the call can’t start, error fires, or the promise rejects when nothing listens for error.
userId, preferHeadphonesForIosDevices, connectionDelay and disableWakeLock are still accepted, so older code keeps working, but they have no effect.
Methods
The call
Microphone and volume
Analytics
getConnectionStats(), getAudioLevels(), getPerformanceMetrics(), getParticipants(), getTrackStats() and getCallAnalytics(). See Read call statistics.
Events
LiveKit objects
The SDK runs on LiveKit. If you build on LiveKit yourself, for example with its React components:Page tools
Each function intools:
How to set up the web tool, read its arguments, return an answer and handle failures: Web tools.
Errors
HamsaApiError extends Error with messageKey, status, params and retryAfter. The keys and what they mean: If a call fails to start.
Exports
HamsaVoiceAgent is also a named export. The SDK also re-exports LiveKit’s Room, RemoteParticipant, RemoteTrack, RemoteTrackPublication, LocalTrack and RpcInvocationData types.
Event handlers are typed: agent.on("callStarted", ({ jobId }) => ...) knows jobId is a string.