@framers/agentos 0.1.175 → 0.1.177

package/dist/voice-pipeline/index.d.ts

@@ -1,39 +1,98 @@
 /**
  * @module @framers/agentos/voice-pipeline
  *
- * Barrel exports for the AgentOS streaming voice pipeline.
+ * Real-time streaming voice pipeline for AgentOS.
  *
- * This module provides all the building blocks needed to assemble a real-time
- * voice conversation system:
+ * Provides a complete, provider-agnostic voice conversation system with
+ * pluggable STT, TTS, endpoint detection, barge-in handling, and transport.
  *
- * - **Types** -- All interfaces and type aliases defining the pipeline's contracts
- *   ({@link AudioFrame}, {@link IStreamTransport}, {@link IEndpointDetector}, etc.).
+ * ## Architecture
  *
- * - **Orchestrator** -- {@link VoicePipelineOrchestrator} is the central state machine
- *   that wires transport, STT, endpoint detection, TTS, and barge-in handling into
- *   a coordinated conversation loop.
+ * ```
+ * Browser Mic → Transport → STT → Endpoint Detector → Agent → TTS → Transport → Browser Speaker
+ *                                                        ↑
+ *                                                  Barge-in Handler
+ * ```
+ *
+ * All components are injected via `VoicePipelineOverrides`, making the pipeline
+ * fully provider-agnostic. Swap Deepgram for ElevenLabs STT, or ElevenLabs for
+ * OpenAI TTS, by changing one line.
+ *
+ * ## Built-in Providers
+ *
+ * **STT (Speech-to-Text):**
+ * - {@link DeepgramStreamingSTT} — WebSocket streaming via Deepgram Nova-2. Lowest latency.
+ * - {@link ElevenLabsStreamingSTT} — Chunked REST via ElevenLabs Scribe. Uses same key as TTS.
  *
- * - **Endpoint Detectors** -- Two strategies for detecting turn boundaries:
- *   - {@link HeuristicEndpointDetector}: Rule-based (punctuation + silence timeout).
- *   - {@link AcousticEndpointDetector}: Purely acoustic (silence-only, no transcript analysis).
+ * **TTS (Text-to-Speech):**
+ * - {@link ElevenLabsStreamingTTS} — WebSocket streaming via ElevenLabs. High quality voices.
  *
- * - **Barge-in Handlers** -- Two strategies for handling user interruptions:
- *   - {@link HardCutBargeinHandler}: Immediate stop above a speech duration threshold.
- *   - {@link SoftFadeBargeinHandler}: Three-tier (ignore/pause/cancel) with configurable fade.
+ * **Endpoint Detection:**
+ * - {@link HeuristicEndpointDetector} — Punctuation + silence timeout. Fast, no model needed.
+ * - {@link AcousticEndpointDetector} — Silence-only, no transcript analysis.
  *
- * - **Transport** -- {@link WebSocketStreamTransport}: WebSocket-based bidirectional
- *   audio/text transport implementing {@link IStreamTransport}.
+ * **Barge-in Handling:**
+ * - {@link HardCutBargeinHandler} — Immediate TTS cancel above speech threshold.
+ * - {@link SoftFadeBargeinHandler} — Three-tier (ignore/pause/cancel) with configurable fade.
  *
- * - **Error** -- {@link VoiceInterruptError}: Typed error for barge-in interruptions.
+ * **Transport:**
+ * - {@link WebSocketStreamTransport} — WebSocket bidirectional audio/text.
+ * - {@link WebRTCStreamTransport} — WebRTC DataChannel transport.
+ *
+ * **Agent Adapter:**
+ * - {@link AgentSessionVoiceAdapter} — Wraps any AgentOS `AgentSession` as `IVoicePipelineAgentSession`.
+ *
+ * ## Usage
  *
- * @example
  * ```typescript
  * import {
  *   VoicePipelineOrchestrator,
  *   HeuristicEndpointDetector,
  *   HardCutBargeinHandler,
  *   WebSocketStreamTransport,
+ *   ElevenLabsStreamingSTT,
+ *   ElevenLabsStreamingTTS,
+ *   AgentSessionVoiceAdapter,
  * } from '../voice-pipeline';
+ * import { agent } from '@framers/agentos';
+ *
+ * // Create agent and voice adapter
+ * const a = agent({ model: 'gpt-4o-mini', instructions: 'You are a voice companion.' });
+ * const session = a.session('voice-1');
+ * const voiceAdapter = new AgentSessionVoiceAdapter(session);
+ *
+ * // Create providers (use whichever API keys you have)
+ * const stt = new ElevenLabsStreamingSTT({ apiKey: process.env.ELEVENLABS_API_KEY! });
+ * const tts = new ElevenLabsStreamingTTS({ apiKey: process.env.ELEVENLABS_API_KEY! });
+ *
+ * // Create and start the pipeline
+ * const orchestrator = new VoicePipelineOrchestrator({
+ *   stt: 'elevenlabs', tts: 'elevenlabs', language: 'en-US',
+ * });
+ *
+ * const pipelineSession = await orchestrator.startSession(transport, voiceAdapter, {
+ *   streamingSTT: stt,
+ *   streamingTTS: tts,
+ *   endpointDetector: new HeuristicEndpointDetector(),
+ *   bargeinHandler: new HardCutBargeinHandler(),
+ * });
+ *
+ * // Listen for state changes (idle → listening → processing → speaking → listening)
+ * pipelineSession.on('state_change', (state) => console.log('Pipeline:', state));
+ * ```
+ *
+ * ## Custom Providers
+ *
+ * Implement {@link IStreamingSTT} and {@link IStreamingTTS} to add any provider:
+ *
+ * ```typescript
+ * class MyCustomSTT implements IStreamingSTT {
+ *   readonly providerId = 'my-custom-stt';
+ *   readonly isStreaming = true;
+ *   async startSession(config?: StreamingSTTConfig): Promise<StreamingSTTSession> {
+ *     // Connect to your STT service, return a session that emits 'transcript' events
+ *   }
+ * }
  * ```
  */
 export * from './types.js';
@@ -45,4 +104,5 @@ export { WebSocketStreamTransport } from './WebSocketStreamTransport.js';
 export { WebRTCStreamTransport, createWebRTCTransport } from './WebRTCStreamTransport.js';
 export { VoicePipelineOrchestrator } from './VoicePipelineOrchestrator.js';
 export { VoiceInterruptError } from './VoiceInterruptError.js';
+export { DeepgramStreamingSTT, type DeepgramStreamingSTTConfig, ElevenLabsStreamingSTT, type ElevenLabsStreamingSTTConfig, ElevenLabsStreamingTTS, type ElevenLabsStreamingTTSConfig, AgentSessionVoiceAdapter, } from './providers/index.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/voice-pipeline/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/voice-pipeline/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqCG;AAIH,cAAc,YAAY,CAAC;AAG3B,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAGrE,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAC3E,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,OAAO,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AAG1F,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAG3E,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/voice-pipeline/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgGG;AAIH,cAAc,YAAY,CAAC;AAG3B,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAGrE,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAC3E,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,OAAO,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AAG1F,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAG3E,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAG/D,OAAO,EACL,oBAAoB,EACpB,KAAK,0BAA0B,EAC/B,sBAAsB,EACtB,KAAK,4BAA4B,EACjC,sBAAsB,EACtB,KAAK,4BAA4B,EACjC,wBAAwB,GACzB,MAAM,sBAAsB,CAAC"}

package/dist/voice-pipeline/index.js

@@ -1,39 +1,98 @@
 /**
  * @module @framers/agentos/voice-pipeline
  *
- * Barrel exports for the AgentOS streaming voice pipeline.
+ * Real-time streaming voice pipeline for AgentOS.
  *
- * This module provides all the building blocks needed to assemble a real-time
- * voice conversation system:
+ * Provides a complete, provider-agnostic voice conversation system with
+ * pluggable STT, TTS, endpoint detection, barge-in handling, and transport.
  *
- * - **Types** -- All interfaces and type aliases defining the pipeline's contracts
- *   ({@link AudioFrame}, {@link IStreamTransport}, {@link IEndpointDetector}, etc.).
+ * ## Architecture
  *
- * - **Orchestrator** -- {@link VoicePipelineOrchestrator} is the central state machine
- *   that wires transport, STT, endpoint detection, TTS, and barge-in handling into
- *   a coordinated conversation loop.
+ * ```
+ * Browser Mic → Transport → STT → Endpoint Detector → Agent → TTS → Transport → Browser Speaker
+ *                                                        ↑
+ *                                                  Barge-in Handler
+ * ```
+ *
+ * All components are injected via `VoicePipelineOverrides`, making the pipeline
+ * fully provider-agnostic. Swap Deepgram for ElevenLabs STT, or ElevenLabs for
+ * OpenAI TTS, by changing one line.
+ *
+ * ## Built-in Providers
+ *
+ * **STT (Speech-to-Text):**
+ * - {@link DeepgramStreamingSTT} — WebSocket streaming via Deepgram Nova-2. Lowest latency.
+ * - {@link ElevenLabsStreamingSTT} — Chunked REST via ElevenLabs Scribe. Uses same key as TTS.
  *
- * - **Endpoint Detectors** -- Two strategies for detecting turn boundaries:
- *   - {@link HeuristicEndpointDetector}: Rule-based (punctuation + silence timeout).
- *   - {@link AcousticEndpointDetector}: Purely acoustic (silence-only, no transcript analysis).
+ * **TTS (Text-to-Speech):**
+ * - {@link ElevenLabsStreamingTTS} — WebSocket streaming via ElevenLabs. High quality voices.
  *
- * - **Barge-in Handlers** -- Two strategies for handling user interruptions:
- *   - {@link HardCutBargeinHandler}: Immediate stop above a speech duration threshold.
- *   - {@link SoftFadeBargeinHandler}: Three-tier (ignore/pause/cancel) with configurable fade.
+ * **Endpoint Detection:**
+ * - {@link HeuristicEndpointDetector} — Punctuation + silence timeout. Fast, no model needed.
+ * - {@link AcousticEndpointDetector} — Silence-only, no transcript analysis.
  *
- * - **Transport** -- {@link WebSocketStreamTransport}: WebSocket-based bidirectional
- *   audio/text transport implementing {@link IStreamTransport}.
+ * **Barge-in Handling:**
+ * - {@link HardCutBargeinHandler} — Immediate TTS cancel above speech threshold.
+ * - {@link SoftFadeBargeinHandler} — Three-tier (ignore/pause/cancel) with configurable fade.
  *
- * - **Error** -- {@link VoiceInterruptError}: Typed error for barge-in interruptions.
+ * **Transport:**
+ * - {@link WebSocketStreamTransport} — WebSocket bidirectional audio/text.
+ * - {@link WebRTCStreamTransport} — WebRTC DataChannel transport.
+ *
+ * **Agent Adapter:**
+ * - {@link AgentSessionVoiceAdapter} — Wraps any AgentOS `AgentSession` as `IVoicePipelineAgentSession`.
+ *
+ * ## Usage
  *
- * @example
  * ```typescript
  * import {
  *   VoicePipelineOrchestrator,
  *   HeuristicEndpointDetector,
  *   HardCutBargeinHandler,
  *   WebSocketStreamTransport,
+ *   ElevenLabsStreamingSTT,
+ *   ElevenLabsStreamingTTS,
+ *   AgentSessionVoiceAdapter,
  * } from '../voice-pipeline/index.js';
+ * import { agent } from '@framers/agentos';
+ *
+ * // Create agent and voice adapter
+ * const a = agent({ model: 'gpt-4o-mini', instructions: 'You are a voice companion.' });
+ * const session = a.session('voice-1');
+ * const voiceAdapter = new AgentSessionVoiceAdapter(session);
+ *
+ * // Create providers (use whichever API keys you have)
+ * const stt = new ElevenLabsStreamingSTT({ apiKey: process.env.ELEVENLABS_API_KEY! });
+ * const tts = new ElevenLabsStreamingTTS({ apiKey: process.env.ELEVENLABS_API_KEY! });
+ *
+ * // Create and start the pipeline
+ * const orchestrator = new VoicePipelineOrchestrator({
+ *   stt: 'elevenlabs', tts: 'elevenlabs', language: 'en-US',
+ * });
+ *
+ * const pipelineSession = await orchestrator.startSession(transport, voiceAdapter, {
+ *   streamingSTT: stt,
+ *   streamingTTS: tts,
+ *   endpointDetector: new HeuristicEndpointDetector(),
+ *   bargeinHandler: new HardCutBargeinHandler(),
+ * });
+ *
+ * // Listen for state changes (idle → listening → processing → speaking → listening)
+ * pipelineSession.on('state_change', (state) => console.log('Pipeline:', state));
+ * ```
+ *
+ * ## Custom Providers
+ *
+ * Implement {@link IStreamingSTT} and {@link IStreamingTTS} to add any provider:
+ *
+ * ```typescript
+ * class MyCustomSTT implements IStreamingSTT {
+ *   readonly providerId = 'my-custom-stt';
+ *   readonly isStreaming = true;
+ *   async startSession(config?: StreamingSTTConfig): Promise<StreamingSTTSession> {
+ *     // Connect to your STT service, return a session that emits 'transcript' events
+ *   }
+ * }
  * ```
  */
 // Re-export all type definitions from the types module.
@@ -53,4 +112,6 @@ export { WebRTCStreamTransport, createWebRTCTransport } from './WebRTCStreamTran
 export { VoicePipelineOrchestrator } from './VoicePipelineOrchestrator.js';
 // Typed error for barge-in interruptions
 export { VoiceInterruptError } from './VoiceInterruptError.js';
+// Streaming provider implementations
+export { DeepgramStreamingSTT, ElevenLabsStreamingSTT, ElevenLabsStreamingTTS, AgentSessionVoiceAdapter, } from './providers/index.js';
 //# sourceMappingURL=index.js.map

package/dist/voice-pipeline/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/voice-pipeline/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqCG;AAEH,wDAAwD;AACxD,8EAA8E;AAC9E,cAAc,YAAY,CAAC;AAE3B,4CAA4C;AAC5C,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAErE,6CAA6C;AAC7C,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAC3E,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,2CAA2C;AAC3C,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,oDAAoD;AACpD,OAAO,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AAE1F,gDAAgD;AAChD,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAE3E,yCAAyC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/voice-pipeline/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgGG;AAEH,wDAAwD;AACxD,8EAA8E;AAC9E,cAAc,YAAY,CAAC;AAE3B,4CAA4C;AAC5C,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAErE,6CAA6C;AAC7C,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAC3E,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,2CAA2C;AAC3C,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,oDAAoD;AACpD,OAAO,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AAE1F,gDAAgD;AAChD,OAAO,EAAE,yBAAyB,EAAE,MAAM,gCAAgC,CAAC;AAE3E,yCAAyC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAE/D,qCAAqC;AACrC,OAAO,EACL,oBAAoB,EAEpB,sBAAsB,EAEtB,sBAAsB,EAEtB,wBAAwB,GACzB,MAAM,sBAAsB,CAAC"}

package/dist/voice-pipeline/providers/AgentSessionVoiceAdapter.d.ts

@@ -0,0 +1,63 @@
+/**
+ * @module voice-pipeline/providers/AgentSessionVoiceAdapter
+ *
+ * Adapts an AgentOS {@link AgentSession} to the {@link IVoicePipelineAgentSession}
+ * interface required by {@link VoicePipelineOrchestrator}.
+ *
+ * The adapter wraps `AgentSession.stream(text)` and yields the resulting
+ * `textStream` (an `AsyncIterable<string>` of token deltas) as the return
+ * value of `sendText()`.
+ *
+ * ## Abort Handling
+ *
+ * The `abort()` method is implemented by setting an internal flag that causes
+ * the `sendText()` iterator to stop yielding tokens. Since `StreamTextResult`
+ * does not expose a native cancellation mechanism, the underlying provider
+ * stream continues but its output is discarded.
+ */
+import type { AgentSession } from '../../api/agent.js';
+import type { IVoicePipelineAgentSession, VoiceTurnMetadata } from '../types.js';
+/**
+ * Wraps an AgentOS `AgentSession` as an `IVoicePipelineAgentSession`.
+ *
+ * @example
+ * ```typescript
+ * import { agent } from '@framers/agentos';
+ * import { AgentSessionVoiceAdapter } from '../../voice-pipeline';
+ *
+ * const a = agent({ model: 'gpt-4o' });
+ * const session = a.session('voice-session-1');
+ * const voiceAdapter = new AgentSessionVoiceAdapter(session);
+ *
+ * // Use with VoicePipelineOrchestrator
+ * orchestrator.startSession(transport, voiceAdapter, overrides);
+ * ```
+ */
+export declare class AgentSessionVoiceAdapter implements IVoicePipelineAgentSession {
+    private readonly session;
+    /** Internal abort flag. Set by `abort()`, checked by the token iterator. */
+    private aborted;
+    constructor(session: AgentSession);
+    /**
+     * Send user text to the agent and yield response tokens as an async iterable.
+     *
+     * The `metadata` parameter carries voice-specific context (speech duration,
+     * endpoint reason, confidence, etc.) that could be injected into the agent's
+     * context for more informed responses. Currently the metadata is not forwarded
+     * to the agent (the AgentSession API doesn't support metadata injection),
+     * but it is available for future enhancement.
+     *
+     * @param text - Transcribed user speech to send to the agent.
+     * @param _metadata - Voice turn metadata (reserved for future use).
+     * @returns An async iterable of response token strings.
+     */
+    sendText(text: string, _metadata: VoiceTurnMetadata): AsyncIterable<string>;
+    /**
+     * Abort the current generation.
+     * Sets an internal flag causing the active `sendText()` iterator to stop
+     * yielding tokens. The underlying LLM stream is not explicitly cancelled
+     * but its output is discarded.
+     */
+    abort(): void;
+}
+//# sourceMappingURL=AgentSessionVoiceAdapter.d.ts.map

package/dist/voice-pipeline/providers/AgentSessionVoiceAdapter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"AgentSessionVoiceAdapter.d.ts","sourceRoot":"","sources":["../../../src/voice-pipeline/providers/AgentSessionVoiceAdapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,KAAK,EAAE,0BAA0B,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAEjF;;;;;;;;;;;;;;;GAeG;AACH,qBAAa,wBAAyB,YAAW,0BAA0B;IAI7D,OAAO,CAAC,QAAQ,CAAC,OAAO;IAHpC,4EAA4E;IAC5E,OAAO,CAAC,OAAO,CAAS;gBAEK,OAAO,EAAE,YAAY;IAElD;;;;;;;;;;;;OAYG;IACI,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,iBAAiB,GAAG,aAAa,CAAC,MAAM,CAAC;IAWlF;;;;;OAKG;IACH,KAAK,IAAI,IAAI;CAGd"}

package/dist/voice-pipeline/providers/AgentSessionVoiceAdapter.js

@@ -0,0 +1,72 @@
+/**
+ * @module voice-pipeline/providers/AgentSessionVoiceAdapter
+ *
+ * Adapts an AgentOS {@link AgentSession} to the {@link IVoicePipelineAgentSession}
+ * interface required by {@link VoicePipelineOrchestrator}.
+ *
+ * The adapter wraps `AgentSession.stream(text)` and yields the resulting
+ * `textStream` (an `AsyncIterable<string>` of token deltas) as the return
+ * value of `sendText()`.
+ *
+ * ## Abort Handling
+ *
+ * The `abort()` method is implemented by setting an internal flag that causes
+ * the `sendText()` iterator to stop yielding tokens. Since `StreamTextResult`
+ * does not expose a native cancellation mechanism, the underlying provider
+ * stream continues but its output is discarded.
+ */
+/**
+ * Wraps an AgentOS `AgentSession` as an `IVoicePipelineAgentSession`.
+ *
+ * @example
+ * ```typescript
+ * import { agent } from '@framers/agentos';
+ * import { AgentSessionVoiceAdapter } from '../../voice-pipeline/index.js';
+ *
+ * const a = agent({ model: 'gpt-4o' });
+ * const session = a.session('voice-session-1');
+ * const voiceAdapter = new AgentSessionVoiceAdapter(session);
+ *
+ * // Use with VoicePipelineOrchestrator
+ * orchestrator.startSession(transport, voiceAdapter, overrides);
+ * ```
+ */
+export class AgentSessionVoiceAdapter {
+    constructor(session) {
+        this.session = session;
+        /** Internal abort flag. Set by `abort()`, checked by the token iterator. */
+        this.aborted = false;
+    }
+    /**
+     * Send user text to the agent and yield response tokens as an async iterable.
+     *
+     * The `metadata` parameter carries voice-specific context (speech duration,
+     * endpoint reason, confidence, etc.) that could be injected into the agent's
+     * context for more informed responses. Currently the metadata is not forwarded
+     * to the agent (the AgentSession API doesn't support metadata injection),
+     * but it is available for future enhancement.
+     *
+     * @param text - Transcribed user speech to send to the agent.
+     * @param _metadata - Voice turn metadata (reserved for future use).
+     * @returns An async iterable of response token strings.
+     */
+    async *sendText(text, _metadata) {
+        this.aborted = false;
+        const result = this.session.stream(text);
+        for await (const token of result.textStream) {
+            if (this.aborted)
+                break;
+            yield token;
+        }
+    }
+    /**
+     * Abort the current generation.
+     * Sets an internal flag causing the active `sendText()` iterator to stop
+     * yielding tokens. The underlying LLM stream is not explicitly cancelled
+     * but its output is discarded.
+     */
+    abort() {
+        this.aborted = true;
+    }
+}
+//# sourceMappingURL=AgentSessionVoiceAdapter.js.map

package/dist/voice-pipeline/providers/AgentSessionVoiceAdapter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"AgentSessionVoiceAdapter.js","sourceRoot":"","sources":["../../../src/voice-pipeline/providers/AgentSessionVoiceAdapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAKH;;;;;;;;;;;;;;;GAeG;AACH,MAAM,OAAO,wBAAwB;IAInC,YAA6B,OAAqB;QAArB,YAAO,GAAP,OAAO,CAAc;QAHlD,4EAA4E;QACpE,YAAO,GAAG,KAAK,CAAC;IAE6B,CAAC;IAEtD;;;;;;;;;;;;OAYG;IACH,KAAK,CAAC,CAAC,QAAQ,CAAC,IAAY,EAAE,SAA4B;QACxD,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC;QAErB,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAEzC,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAC5C,IAAI,IAAI,CAAC,OAAO;gBAAE,MAAM;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED;;;;;OAKG;IACH,KAAK;QACH,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;IACtB,CAAC;CACF"}

package/dist/voice-pipeline/providers/ElevenLabsStreamingSTT.d.ts

@@ -0,0 +1,70 @@
+/**
+ * @module voice-pipeline/providers/ElevenLabsStreamingSTT
+ *
+ * Streaming speech-to-text adapter for ElevenLabs' WebSocket STT API.
+ * Implements {@link IStreamingSTT} / {@link StreamingSTTSession} for the
+ * voice pipeline orchestrator.
+ *
+ * ## ElevenLabs STT WebSocket Protocol
+ *
+ * - **Endpoint:** `wss://api.elevenlabs.io/v1/speech-to-text/stream`
+ * - **Authentication:** `xi-api-key` header on upgrade
+ * - **Inbound (client → ElevenLabs):** Binary PCM frames (16-bit signed LE, 16kHz mono)
+ * - **Outbound (ElevenLabs → client):** JSON transcript results
+ * - **Close:** Send JSON `{ "type": "close_stream" }` to finalize
+ *
+ * ## Fallback: Chunked REST
+ *
+ * If the WebSocket endpoint is unavailable or errors, this adapter falls back
+ * to a chunked REST approach: accumulates audio into ~2s chunks and POSTs each
+ * to `/v1/speech-to-text` for batch transcription. This provides near-realtime
+ * results (2s latency per chunk) using only the REST API.
+ *
+ * @see https://elevenlabs.io/docs/api-reference/speech-to-text
+ */
+import type { IStreamingSTT, StreamingSTTSession, StreamingSTTConfig } from '../types.js';
+/**
+ * Configuration for the {@link ElevenLabsStreamingSTT} provider.
+ */
+export interface ElevenLabsStreamingSTTConfig {
+    /** ElevenLabs API key. */
+    apiKey: string;
+    /**
+     * Base URL for the ElevenLabs API.
+     * @default 'https://api.elevenlabs.io/v1'
+     */
+    baseUrl?: string;
+    /**
+     * STT model to use.
+     * @default 'scribe_v1'
+     */
+    model?: string;
+}
+/**
+ * Streaming STT provider using ElevenLabs' Speech-to-Text API.
+ *
+ * Uses chunked REST transcription (2-second audio windows) to provide
+ * near-realtime STT with the same ElevenLabs API key used for TTS.
+ * No separate Deepgram key required.
+ *
+ * @example
+ * ```typescript
+ * const stt = new ElevenLabsStreamingSTT({
+ *   apiKey: process.env.ELEVENLABS_API_KEY!,
+ * });
+ * const session = await stt.startSession({ language: 'en' });
+ * session.on('transcript', (event) => console.log(event.text));
+ * ```
+ */
+export declare class ElevenLabsStreamingSTT implements IStreamingSTT {
+    private readonly config;
+    readonly providerId = "elevenlabs-streaming-stt";
+    readonly isStreaming = true;
+    constructor(config: ElevenLabsStreamingSTTConfig);
+    /**
+     * Create a new STT session. Uses chunked REST calls to ElevenLabs'
+     * batch STT endpoint for near-realtime transcription.
+     */
+    startSession(config?: StreamingSTTConfig): Promise<StreamingSTTSession>;
+}
+//# sourceMappingURL=ElevenLabsStreamingSTT.d.ts.map

package/dist/voice-pipeline/providers/ElevenLabsStreamingSTT.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ElevenLabsStreamingSTT.d.ts","sourceRoot":"","sources":["../../../src/voice-pipeline/providers/ElevenLabsStreamingSTT.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAIH,OAAO,KAAK,EACV,aAAa,EACb,mBAAmB,EACnB,kBAAkB,EAInB,MAAM,aAAa,CAAC;AAMrB;;GAEG;AACH,MAAM,WAAW,4BAA4B;IAC3C,0BAA0B;IAC1B,MAAM,EAAE,MAAM,CAAC;IAEf;;;OAGG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB;;;OAGG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAoQD;;;;;;;;;;;;;;;GAeG;AACH,qBAAa,sBAAuB,YAAW,aAAa;IAI9C,OAAO,CAAC,QAAQ,CAAC,MAAM;IAHnC,QAAQ,CAAC,UAAU,8BAA8B;IACjD,QAAQ,CAAC,WAAW,QAAQ;gBAEC,MAAM,EAAE,4BAA4B;IAEjE;;;OAGG;IACG,YAAY,CAAC,MAAM,CAAC,EAAE,kBAAkB,GAAG,OAAO,CAAC,mBAAmB,CAAC;CAG9E"}