# Changelog All notable changes to `@audin.ai/operator-sdk` are documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). Because the SDK talks to the Audin operator service over a versioned wire protocol (currently **v1**), an **incompatible change to that protocol is a breaking change** and triggers a MAJOR version bump. ## [Unreleased] ### Changed - License changed from `UNLICENSED` to **MIT** ahead of the first public npm release. The SDK is now distributable under the MIT terms (see `LICENSE`). ## [0.4.0] - 2026-06-10 ### Added - **`call_ended` presence notification handling.** The platform now pushes `{ type: "call_ended", callSid, reason }` on the presence channel whenever a call terminates — remote hangup, no answer, busy — INCLUDING the cases where the audio leg never opened (e.g. an outbound the far end never answered), which the audio-WS close alone could never signal. The SDK ends the matching active call, tears the audio bridge down, and emits a single `callEnded` with `endReason` mapped to `remote_hangup` / `no_answer` / `failed`. Additive protocol change (still **v1**): older SDKs simply ignore the new message and keep relying on the audio-WS close. ### Fixed - Internal call teardown paths (`goOffline` with an active call, audio start failure) could emit `callEnded` TWICE for the same call — once from the bridge's synchronous `onClosed` and once from the teardown helper — and the bridge's `remote_hangup` reason would override the intended one (`offline` / `failed`). Teardown now clears the call refs before closing the bridge, so exactly one `callEnded` is emitted, with the correct reason. ## [0.3.0] - 2026-06-10 ### Added - **`setAudioConstraints(constraints)`** on `AudinOperator` — switch the microphone input device (or any `MediaTrackConstraints`) at runtime, e.g. `op.setAudioConstraints({ deviceId: { exact: id }, echoCancellation: true })`. The new constraints are read when the next call's audio leg opens, so the change takes effect from the NEXT `dial()`/accepted inbound call; a call already in progress keeps the device its bridge opened with (the microphone is not re-opened mid-call). ## [0.2.1] - 2026-06-10 ### Fixed - Capture/playback frozen after first audio block — resample read-head overshoot caused a `RangeError` in the worklet. The capture resampler's fractional read head could advance past the end of the accumulated buffer, so on the next `process()` the buffer-grow step computed a negative offset and threw, stopping the processor after the very first block and muting both directions. The read head is now clamped to the buffer length so it never produces a negative offset. Also added an `onprocessorerror` handler so any future worklet error is logged instead of failing silently. ## [0.2.0] - 2026-06-10 ### Added - **`listPhoneNumbers()`** on `AudinOperator` — fetch the phone numbers the account owns (`{ id, phoneNumber, displayName }[]`) directly via the SDK using the same short-lived session token as the WebSockets. The Account API Key never enters the browser. On a `401` the cached token is dropped and the request is retried once with a fresh token; a persistent `401` throws an `OperatorRequestError` with code `UNAUTHORIZED`. New exports: `OperatorPhoneNumber` (type) and `OperatorRequestError` (class). - **Token manager (internal)** — session tokens are now obtained through a central cache shared by the REST helper, the presence WebSocket and the audio WebSocket. A valid cached token is reused (using its `expiresAt` when present, with a safety skew, or a conservative TTL otherwise) so a `listPhoneNumbers()` → `goOnline()` → `dial()` sequence mints a single token. Concurrent callers share one in-flight `getToken()`. The cache is invalidated on auth failures (REST `401` / auth-related WebSocket close) so the next use re-mints. ## [0.1.0] - 2026-06-09 Initial public release. ### Added - **Presence client** — connect/disconnect an operator presence channel, go online on a set of phone numbers and receive availability confirmations, with automatic reconnection (configurable backoff) and heartbeats. - **Audio client** — microphone capture and full-duplex audio over the call's audio channel, with mute/unmute. - **AudioWorklet μ-law pipeline** — capture-side downsampling to 8 kHz + G.711 μ-law encoding, playback-side μ-law decoding + upsampling, run off the main thread. Codec and resampling helpers (`encodeMuLaw`, `decodeMuLaw`, `resampleLinear`, …) are also exported for advanced integrators. - **Outbound calls** — `dial(to, { callerId })` placing an outbound call. - **Inbound calls** — `incomingCall` events with `accept()` / `reject()`. - **Active-call controls** — `mute(on)` and `hangup()`. - **Single-active-call concurrency (MVP)** — while a call is live, new inbound offers are auto-declined and `dial()` rejects. - **Typed event API** — `on` / `off` / `once` for `presenceStateChanged`, `availabilityChanged`, `incomingCall`, `callStarted`, `callEnded`, `error`. - Ships as **ESM** (`dist/audin-operator-sdk.js`) and **UMD** (`dist/audin-operator-sdk.umd.cjs`) with full **TypeScript declarations** (`dist/index.d.ts`). Zero runtime dependencies. ### Known limitations - **`sendDtmf` is not yet supported end-to-end.** The digit is validated and emitted as a control message, but the server does not yet forward the tones onto the telephone network, so the far end will not hear them. The method and its wire message are kept so that enabling it server-side in a future release needs no SDK change. [Unreleased]: https://bitbucket.org/thecovesrl/audin-ai/branches/compare/main%0Av0.3.0 [0.3.0]: https://bitbucket.org/thecovesrl/audin-ai/branches/compare/v0.3.0%0Av0.2.1 [0.2.1]: https://bitbucket.org/thecovesrl/audin-ai/branches/compare/v0.2.1%0Av0.2.0 [0.2.0]: https://bitbucket.org/thecovesrl/audin-ai/branches/compare/v0.2.0%0Av0.1.0 [0.1.0]: https://bitbucket.org/thecovesrl/audin-ai/src/v0.1.0/audin-operator-sdk