A voice engine for
real-time agents.

// Cascade: STT -> your reasoner -> TTS
import { VoiceAgentSession } from '@kuralle-syrinx/core';
import { DeepgramSTTPlugin } from '@kuralle-syrinx/deepgram';
import { CartesiaTTSPlugin } from '@kuralle-syrinx/cartesia';
import { ReasoningBridge, fromStreamText } from '@kuralle-syrinx/aisdk';

const session = new VoiceAgentSession({
  plugins: {
    stt: { api_key: DEEPGRAM_API_KEY, model: 'nova-3', sample_rate: 16000 },
    bridge: {},
    tts: { api_key: CARTESIA_API_KEY, voice_id: CARTESIA_VOICE_ID },
  },
  endpointingOwner: 'provider_stt',
});

session.registerPlugin('stt', new DeepgramSTTPlugin());
session.registerPlugin('bridge', new ReasoningBridge(fromStreamText({ model, system })));
session.registerPlugin('tts', new CartesiaTTSPlugin());

// Native realtime: one speech-to-speech model
import { RealtimeBridge, fromOpenAIRealtime } from '@kuralle-syrinx/realtime';

session.registerPlugin(
  'realtime',
  new RealtimeBridge(fromOpenAIRealtime({ apiKey, socketFactory })),
);

// Same tools, resume, and observability as the cascade —
// swap OpenAI Realtime for Gemini Live with one import.

// Ship on Cloudflare Workers — one hibernatable
// Durable Object per conversation, same engine.
import { Agent } from 'agents';
import { withVoice } from '@kuralle-syrinx/cf-agents';

export class VoiceConversation extends withVoice(Agent, {
  pipeline: {
    kind: 'cascaded',
    stt: (env) => ({ plugin: new DeepgramSTTPlugin(createWorkersSocket), config }),
    tts: (env) => ({ plugin: new CartesiaTTSPlugin(createWorkersSocket), config }),
  },
  reasoner,
}) {}