@recombine-ai/platform 0.2.6-test.8 → 0.3.0

package/build/call-control/schemas.generated.js

@@ -0,0 +1,281 @@
+import { z } from 'zod';
+export const configSchema = z.object({
+    agent_background_audio_volume: z.number().default(0.2),
+    agent_background_audio_autopause: z.boolean().default(false),
+    session_turn_detection: z
+        .enum(['stt', 'vad', 'model_multi', 'model_en'])
+        .describe("Strategy for deciding when the user has finished speaking. 'stt' relies on speech-to-text end-of-utterance cues, 'vad' relies on Voice Activity Detection start/stop cues, 'model_multi' and 'model_en' use a turn detector model")
+        .default('model_en'),
+    session_allow_interruptions: z
+        .boolean()
+        .describe('Whether the user can interrupt the agent mid-utterance')
+        .default(true),
+    agent_dynamic_interruption_escalation: z
+        .boolean()
+        .describe('Whether to dynamically escalate interruption threshold if user does not interrupt within the defined delay after TTS starts')
+        .default(false),
+    session_min_interruption_words: z
+        .number()
+        .int()
+        .describe('Minimum number of transcribed words to consider an interruption (only used when STT is enabled)')
+        .default(1),
+    session_min_interruption_duration: z
+        .number()
+        .describe('Minimum speech length in seconds to register as an interruption')
+        .default(0.1),
+    session_min_endpointing_delay: z
+        .number()
+        .describe("Minimum time in seconds since the last detected speech before the agent declares the user's turn complete. In VAD mode behaves like max(VAD silence, min_endpointing_delay); in STT mode applied after the STT end-of-speech signal, so it can be additive with the STT provider's endpointing delay")
+        .default(0.1),
+    session_max_endpointing_delay: z
+        .number()
+        .describe('Maximum time in seconds the agent will wait before terminating the turn')
+        .default(3),
+    session_preemptive_generation: z
+        .boolean()
+        .describe('Whether to speculatively begin LLM and TTS requests before end-of-turn is detected. Reduces response latency by overlapping model inference with user audio, but may incur extra compute if the user interrupts mid-utterance')
+        .default(true),
+    vad_min_speech_duration: z
+        .number()
+        .describe('Minimum duration of speech in seconds to start a new speech chunk')
+        .default(0.02),
+    vad_min_silence_duration: z
+        .number()
+        .describe('At the end of each speech, wait this duration in seconds before ending the speech')
+        .default(1),
+    vad_prefix_padding_duration: z
+        .number()
+        .describe('Duration in seconds of padding to add to the beginning of each speech chunk')
+        .default(0.5),
+    vad_max_buffered_speech: z
+        .number()
+        .describe('Maximum duration of speech to keep in the buffer, in seconds')
+        .default(180),
+    vad_activation_threshold: z
+        .number()
+        .describe('Threshold to consider a frame as speech')
+        .default(0.1),
+    vad_deactivation_threshold: z
+        .union([z.number(), z.null()])
+        .describe("Negative threshold (noise or exit threshold). If model's current state is SPEECH, values below this are considered as NON-SPEECH. Default is max(activation_threshold - 0.15, 0.01)")
+        .default(null),
+    vad_sample_rate: z
+        .union([z.literal(8000), z.literal(16000)])
+        .describe('Sample rate for VAD inference (only 8KHz and 16KHz are supported)')
+        .default(8000),
+    stt_api_version: z
+        .union([z.literal(1), z.literal(2)])
+        .describe('Deepgram API version. Version 1 is the default with rich configuration options. Version 2 uses the Flux model family with end-of-turn detection (eot_threshold, eot_timeout_ms, eager_eot_threshold)')
+        .default(1),
+    stt_model: z
+        .union([
+        z.literal('flux-general-en'),
+        z.enum([
+            'nova-general',
+            'nova-phonecall',
+            'nova-meeting',
+            'nova-2-general',
+            'nova-2-meeting',
+            'nova-2-phonecall',
+            'nova-2-finance',
+            'nova-2-conversationalai',
+            'nova-2-voicemail',
+            'nova-2-video',
+            'nova-2-medical',
+            'nova-2-drivethru',
+            'nova-2-automotive',
+            'nova-3',
+            'nova-3-general',
+            'nova-3-medical',
+            'enhanced-general',
+            'enhanced-meeting',
+            'enhanced-phonecall',
+            'enhanced-finance',
+            'base',
+            'meeting',
+            'phonecall',
+            'finance',
+            'conversationalai',
+            'voicemail',
+            'video',
+            'whisper-tiny',
+            'whisper-base',
+            'whisper-small',
+            'whisper-medium',
+            'whisper-large',
+        ]),
+    ])
+        .describe('The Deepgram model to use for speech recognition')
+        .default('nova-3'),
+    stt_language: z
+        .enum([
+        'zh',
+        'zh-CN',
+        'zh-TW',
+        'da',
+        'nl',
+        'en',
+        'en-US',
+        'en-AU',
+        'en-GB',
+        'en-NZ',
+        'en-IN',
+        'fr',
+        'fr-CA',
+        'de',
+        'hi',
+        'hi-Latn',
+        'pt',
+        'pt-BR',
+        'es',
+        'es-419',
+        'it',
+        'ja',
+        'ko',
+        'no',
+        'pl',
+        'es-LATAM',
+        'sv',
+        'ta',
+        'taq',
+        'uk',
+        'tr',
+        'id',
+        'ru',
+        'th',
+        'multi',
+    ])
+        .describe('The language code for recognition')
+        .default('en'),
+    stt_detect_language: z
+        .boolean()
+        .describe('Whether to enable automatic language detection')
+        .default(false),
+    stt_interim_results: z
+        .boolean()
+        .describe('Whether to return interim (non-final) transcription results')
+        .default(false),
+    stt_punctuate: z
+        .boolean()
+        .describe('Whether to add punctuations to the transcription. Turn detector will work better with punctuations')
+        .default(true),
+    stt_smart_format: z
+        .boolean()
+        .describe('Whether to apply smart formatting to numbers, dates, etc.')
+        .default(false),
+    stt_sample_rate: z.number().int().describe('The sample rate of the audio in Hz').default(8000),
+    stt_no_delay: z
+        .boolean()
+        .describe('When smart_format is used, ensures it does not wait for sequence to be complete before returning results')
+        .default(true),
+    stt_endpointing_ms: z
+        .number()
+        .int()
+        .describe('Time in milliseconds of silence to consider end of speech. Set to 0 to disable')
+        .default(25),
+    stt_filler_words: z
+        .boolean()
+        .describe('Whether to include filler words (um, uh, etc.) in transcription.')
+        .default(true),
+    stt_keyterms: z
+        .array(z.string())
+        .describe('Key terms to improve recognition accuracy. Only supported by Nova-3 models')
+        .default([]),
+    stt_tags: z
+        .array(z.string())
+        .describe('List of tags to add to the requests for usage reporting.')
+        .default([]),
+    stt_profanity_filter: z
+        .boolean()
+        .describe('Whether to filter profanity from the transcription.')
+        .default(false),
+    stt_eager_eot_threshold: z
+        .union([z.number(), z.null()])
+        .describe('Threshold for eager end-of-turn to enable preemptive generation. Disabled by default. Set to 0.3-0.9 to enable. Only used when stt_api_version is 2')
+        .default(null),
+    stt_eot_threshold: z
+        .number()
+        .describe('Threshold for end-of-turn detection. Only used when stt_api_version is 2')
+        .default(0.7),
+    stt_eot_timeout_ms: z
+        .number()
+        .int()
+        .describe('Timeout in milliseconds for end-of-turn detection. Only used when stt_api_version is 2')
+        .default(3000),
+    tts_voice_id: z
+        .string()
+        .describe('ElevenLabs voice ID (default `bIHbv24MWmeRgasZH58o` used if none set)')
+        .default('bIHbv24MWmeRgasZH58o'),
+    tts_voice_similarity_boost: z
+        .number()
+        .describe('Voice similarity boost [0.0 - 1.0]')
+        .default(0.6),
+    tts_voice_stability: z.number().describe('Voice stability [0.0 - 1.0]').default(0.95),
+    tts_voice_speed: z.number().describe('Voice speed [0.8 - 1.2]').default(1),
+    tts_voice_style: z.number().describe('Voice style exaggeration [0.0 - 1.0]').default(0.1),
+    tts_voice_use_speaker_boost: z
+        .boolean()
+        .describe('Whether to use speaker boost for the voice')
+        .default(false),
+    tts_model: z
+        .enum([
+        'eleven_monolingual_v1',
+        'eleven_multilingual_v1',
+        'eleven_multilingual_v2',
+        'eleven_turbo_v2',
+        'eleven_turbo_v2_5',
+        'eleven_flash_v2_5',
+        'eleven_flash_v2',
+        'eleven_v3',
+    ])
+        .describe('ElevenLabs model to use')
+        .default('eleven_turbo_v2'),
+    tts_inactivity_timeout: z
+        .number()
+        .int()
+        .describe('Inactivity timeout in seconds for the websocket connection')
+        .default(300),
+    tts_enable_ssml_parsing: z
+        .boolean()
+        .describe('Enable SSML parsing for input text')
+        .default(true),
+    tts_streaming_latency: z
+        .number()
+        .int()
+        .describe('Optimize for streaming latency, 0 to disable, 4 for max latency optimizations (deprecated)')
+        .default(3),
+    tts_auto_mode: z
+        .boolean()
+        .describe('Reduces latency by disabling chunk schedule and buffers. Sentence tokenizer will be used to synthesize one sentence at a time')
+        .default(true),
+    tts_encoding: z
+        .enum([
+        'mp3_22050_32',
+        'mp3_24000_48',
+        'mp3_44100',
+        'mp3_44100_32',
+        'mp3_44100_64',
+        'mp3_44100_96',
+        'mp3_44100_128',
+        'mp3_44100_192',
+        'opus_48000_32',
+        'opus_48000_64',
+        'opus_48000_96',
+        'opus_48000_128',
+        'opus_48000_192',
+        'pcm_8000',
+        'pcm_16000',
+        'pcm_22050',
+        'pcm_24000',
+        'pcm_32000',
+        'pcm_44100',
+        'pcm_48000',
+    ])
+        .describe('Audio encoding format for TTS output')
+        .default('mp3_44100_128'),
+    tts_chunk_length_schedule: z
+        .array(z.number().int())
+        .describe('Schedule for chunk lengths, ranging 50...500')
+        .default([120, 160, 250, 290]),
+    tts_language: z.string().describe('Language code for the TTS model').default('en'),
+});

package/build/call-control/sip-setup-lock.d.ts

@@ -0,0 +1,6 @@
+import type { Logger } from './types';
+export default function createSipSetupLock(url: string, key: string, secret: string, logger?: Logger): {
+    acquire: () => Promise<void>;
+    release: () => Promise<void>;
+};
+//# sourceMappingURL=sip-setup-lock.d.ts.map

package/build/call-control/sip-setup-lock.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sip-setup-lock.d.ts","sourceRoot":"","sources":["../../src/call-control/sip-setup-lock.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,SAAS,CAAA;AASrC,MAAM,CAAC,OAAO,UAAU,kBAAkB,CACtC,GAAG,EAAE,MAAM,EACX,GAAG,EAAE,MAAM,EACX,MAAM,EAAE,MAAM,EACd,MAAM,GAAE,MAAgB,GACzB;IAAE,OAAO,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;IAAC,OAAO,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;CAAE,CAkDhE"}

package/build/call-control/sip-setup-lock.js

@@ -0,0 +1,60 @@
+import { randomUUID } from 'node:crypto';
+import { RoomServiceClient } from 'livekit-server-sdk';
+import { CallControlError } from './utils';
+const LOCK_ROOM_NAME = '__sip_setup_lock__';
+const LOCK_ROOM_EMPTY_TIMEOUT_S = 60;
+const LOCK_ACQUIRE_TIMEOUT_MS = 30_000;
+const LOCK_RETRY_INTERVAL_MS = 500;
+const LOCK_REQUEST_TIMEOUT_S = 5;
+export default function createSipSetupLock(url, key, secret, logger = console) {
+    const roomService = new RoomServiceClient(url, key, secret, {
+        requestTimeout: LOCK_REQUEST_TIMEOUT_S,
+    });
+    let lockId;
+    return {
+        async acquire() {
+            lockId = randomUUID();
+            const startTime = Date.now();
+            while (Date.now() - startTime < LOCK_ACQUIRE_TIMEOUT_MS) {
+                try {
+                    const room = await roomService.createRoom({
+                        name: LOCK_ROOM_NAME,
+                        emptyTimeout: LOCK_ROOM_EMPTY_TIMEOUT_S,
+                        agents: [],
+                        metadata: lockId,
+                    });
+                    if (room.metadata === lockId) {
+                        logger.info('SIP setup lock acquired');
+                        return;
+                    }
+                    logger.info('SIP setup lock held by another instance, waiting...', {
+                        holder: room.metadata,
+                    });
+                }
+                catch (err) {
+                    logger.warn('SIP setup lock acquire attempt failed, retrying...', { err });
+                }
+                await new Promise((resolve) => setTimeout(resolve, LOCK_RETRY_INTERVAL_MS));
+            }
+            lockId = undefined;
+            throw new CallControlError(`Failed to acquire SIP setup lock within ${LOCK_ACQUIRE_TIMEOUT_MS}ms`);
+        },
+        async release() {
+            try {
+                await roomService.deleteRoom(LOCK_ROOM_NAME);
+                logger.info('SIP setup lock released');
+            }
+            catch (err) {
+                if (err instanceof Error && err.message.includes('not found')) {
+                    logger.debug('SIP setup lock already released');
+                    return;
+                }
+                logger.error('Failed to release SIP setup lock', { err });
+                throw err;
+            }
+            finally {
+                lockId = undefined;
+            }
+        },
+    };
+}

package/build/call-control/sip.d.ts

@@ -0,0 +1,41 @@
+import { SIPDispatchRuleInfo as SipDispatchRule, SIPInboundTrunkInfo as SipInboundTrunk, SIPOutboundTrunkInfo as SipOutboundTrunk } from 'livekit-server-sdk';
+import type { Logger, SipManagerOptions, SipResourceMap } from './types';
+import { CallControlError } from './utils';
+type SipSetupReturn = {
+    agentName: string;
+    inboundTrunk: SipInboundTrunk | undefined;
+    transferTrunk: SipOutboundTrunk | undefined;
+    outboundTrunk: SipOutboundTrunk | undefined;
+    dispatchRule: SipDispatchRule | undefined;
+};
+export declare class SipManager {
+    agentName: string;
+    private client;
+    private lock;
+    private options;
+    private trunksConfigHash;
+    private logger;
+    constructor(options: SipManagerOptions, logger?: Logger, debugMode?: boolean);
+    private get trunksConfig();
+    private get trunksConfigDefaultMeta();
+    private validateOptions;
+    setup(withCleanup?: boolean): Promise<SipSetupReturn>;
+    getTransferTrunkId(): Promise<string | undefined>;
+    private createInboundTrunk;
+    private createOutboundTrunk;
+    private createDispatchRule;
+    private resolveNumberConflicts;
+    private isOwnResource;
+    private isResourceSetComplete;
+    private cleanup;
+    private deleteAllResources;
+    getAllResources(): Promise<SipResourceMap>;
+    private deleteTrunk;
+    private deleteDispatchRule;
+}
+export declare class SipError extends CallControlError {
+    constructor(message: string, cause?: unknown);
+}
+export declare function populateOptionsFromEnv(options?: Partial<SipManagerOptions>, parse?: (text: string, reviver?: (this: any, key: string, value: any) => any) => any): Partial<SipManagerOptions>;
+export {};
+//# sourceMappingURL=sip.d.ts.map

package/build/call-control/sip.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sip.d.ts","sourceRoot":"","sources":["../../src/call-control/sip.ts"],"names":[],"mappings":"AAAA,OAAO,EAOH,mBAAmB,IAAI,eAAe,EACtC,mBAAmB,IAAI,eAAe,EACtC,oBAAoB,IAAI,gBAAgB,EAC3C,MAAM,oBAAoB,CAAA;AAG3B,OAAO,KAAK,EACR,MAAM,EAEN,iBAAiB,EAEjB,cAAc,EAEjB,MAAM,SAAS,CAAA;AAEhB,OAAO,EACH,gBAAgB,EAKnB,MAAM,SAAS,CAAA;AAYhB,KAAK,cAAc,GAAG;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,eAAe,GAAG,SAAS,CAAA;IACzC,aAAa,EAAE,gBAAgB,GAAG,SAAS,CAAA;IAC3C,aAAa,EAAE,gBAAgB,GAAG,SAAS,CAAA;IAC3C,YAAY,EAAE,eAAe,GAAG,SAAS,CAAA;CAC5C,CAAA;AAED,qBAAa,UAAU;IACZ,SAAS,EAAE,MAAM,CAAA;IACxB,OAAO,CAAC,MAAM,CAAW;IACzB,OAAO,CAAC,IAAI,CAAuC;IACnD,OAAO,CAAC,OAAO,CAAmB;IAClC,OAAO,CAAC,gBAAgB,CAAQ;IAChC,OAAO,CAAC,MAAM,CAAQ;gBAGlB,OAAO,EAAE,iBAAiB,EAC1B,MAAM,GAAE,MAAgB,EACxB,SAAS,GAAE,OAAwC;IAiBvD,OAAO,KAAK,YAAY,GAEvB;IAED,OAAO,KAAK,uBAAuB,GAQlC;IAED,OAAO,CAAC,eAAe;IAoBV,KAAK,CAAC,WAAW,GAAE,OAAc,GAAG,OAAO,CAAC,cAAc,CAAC;IAsE3D,kBAAkB,IAAI,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC;YAQhD,kBAAkB;YAyBlB,mBAAmB;YAgCnB,kBAAkB;YAuClB,sBAAsB;IA+CpC,OAAO,CAAC,aAAa,CAEpB;IAED,OAAO,CAAC,qBAAqB;YAef,OAAO;YAIP,kBAAkB;IAsFnB,eAAe,IAAI,OAAO,CAAC,cAAc,CAAC;YASzC,WAAW;YAcX,kBAAkB;CAanC;AAcD,qBAAa,QAAS,SAAQ,gBAAgB;gBAC9B,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,OAAO;CAI/C;AAED,wBAAgB,sBAAsB,CAClC,OAAO,GAAE,OAAO,CAAC,iBAAiB,CAAM,EACxC,KAAK,+EAAa,GACnB,OAAO,CAAC,iBAAiB,CAAC,CAuC5B"}