v5.8.0

5.8.0 (2026-08-12)

Flux TTS streaming controls, Listen v2 redaction, and an agent provider back-compat fix.

Features

• Speak v2 (Flux TTS streaming, /v2/speak): barge-in via sendInterrupt() (SpeakV2Interrupt, optional playback_offset), answered by SpeakV2SpeechInterrupted with text_spoken / text_remaining; mid-session Configure via sendConfigure() with ConfigureSuccess / ConfigureFailure responses; new speed (0.85–1.15 in 0.05 increments) and expressivity (-2–2; 0 is the voice's nominal delivery) connect params. Inline pause and pronunciation controls are not applied at launch — they are stripped before synthesis, the related warning codes are reserved and not currently emitted, and controls_applied counts are currently 0; support is coming soon. (#532) (e851496)
• Listen v2: ListenV2Redact (numbers, aggressive_numbers). (#532) (e851496)

Bug Fixes

• agent: model the AgentV1UpdateListen provider as a merged shape rather than a union, preserving back-compat for existing callers (the regenerated union broke reads of V2-only fields and required version). Compile-compat only; the wire payload is unchanged. (a65d1aa, a6e3c79)

Compatibility

• No optional → required field changes on request types.
• Two existing fields widen on the read path: Deepgram.version ("v1" → string) and Google.version ("v1beta" → open Google.Version). Safe for callers setting the field; a consumer assigning them into a narrowed literal binding must widen the annotation.
• SpeakV2SpeechMetadata.ControlsApplied / SpeakV2SpeechInterrupted gain a required breaks_applied counter — server-emitted (read-only), so it is safe on the read path.

v5.7.0

5.7.0 (2026-07-22)

Features

• regen: listen v2 numerals, multilingual Aura-2 voices, and streaming close() fix (#522) (7f0b9e0)

Bug Fixes

• agent: re-add deprecated stt_latency to AgentV1LatencyReport (back-compat shim) (007c7e4)
• socket: make streaming close() idempotent to prevent close-handler recursion (bf3d1df)

What's in this release

SDK regeneration for 2026-07-20 (spec ff8fd2b). No breaking changes.

New API surface (Fern-owned, additive)

• listen.v2 numerals — new numerals param + ListenV2Numerals type. Connection-time only. Verified end-to-end (serializes onto the connect URL; a TTS'd digits phrase transcribes as 5 5 5 1 2 3 4 … 42 … 3rd with numerals: true vs spelled-out with false).
• New Aura-2 multilingual TTS voices — ~40 voices (es / de / nl / fr / it / ja) on SpeakV1Model / AudioGenerateRequestModel. Additive only.

Bug fix — streaming close() recursion (bf3d1df)

• The generated Socket.close() synchronously re-fires the registered close handler via handleClose(); a close handler that calls close() again (idiomatic cleanup) recursed infinitely → RangeError: Maximum call stack size exceeded. All 5 streaming sockets were affected (agent.v1, listen.v1, listen.v2, speak.v1, speak.v2); pre-existing on main.
• Fixed at the wrapper layer (CustomClient.ts, already frozen) with a shared WeakSet guard — no generated files frozen. Regression test drives the public createConnection path for all 5 services.

Bug fix — stt_latency back-compat shim (007c7e4)

• The spec removed stt_latency from AgentV1LatencyReport (docs #1006). Since it's a server-emitted (read-only) message, stt_latency?: number is re-added by hand as a deprecated read-side field (frozen in .fernignore): report.stt_latency keeps resolving (to undefined) instead of breaking existing readers at compile time — no request/wire impact. Mirrors the Python SDK's read-side shim (deepgram-python-sdk#746).
• Both shims work around generator output and can be dropped once corrected upstream.

Validation — all green

make build ✅ · make test ✅ (unit 580/580, wire 139/139) · make typecheck-tests ✅ · biome lint ✅ · biome format ✅ · 27/27 non-management examples run clean against the live API.

v5.6.0

chore(main): release 5.6.0 (#518)

Release PR for **5.6.0** — [compare
v5.5.0...v5.6.0](https://github.com/deepgram/deepgram-js-sdk/compare/v5.5.0...v5.6.0)

## What's in 5.6.0

- **[#515](https://github.com/deepgram/deepgram-js-sdk/pull/515) — Flux
TTS streaming (Speak V2) + agent listen reconfigure**
- **Speak V2 streaming (WebSocket)** — `client.speak.v2.connect(...)` →
`SpeakV2Socket`: new `speak/v2` package and `SpeakV2*` types, wired into
`DeepgramClient` with a binary-aware socket (mirrors the `speak.v1`
pair) so audio frames aren't JSON-parsed away.
- **Agent mid-session listen reconfigure** — `AgentV1UpdateListen` /
`AgentV1UpdateListenListen` messages with the `AgentV1ListenUpdated`
acknowledgement.
- **Flux end-of-turn tuning** — new `DeepgramListenProviderV2` fields
`eot_threshold`, `eager_eot_threshold`, `eot_timeout_ms`.
- **Reliability** — unknown/future server frames pass through untouched
(forward-compat); `ListenV2CloseStream` type corrected to only
`CloseStream`.

- **[#519](https://github.com/deepgram/deepgram-js-sdk/pull/519) — Flux
TTS batch (REST) + agent latency report**
- **Speak V2 batch (REST)** — `client.speak.v2.audio.generate(...)`: the
REST companion to streaming. New `SpeakV2Request` / `SpeakV2Response` /
`SpeakV2AcceptedResponse`, `AudioGenerateRequest*` types, and the
`./speak/v2/audio` export.
- `sample_rate` / `bit_rate` serialize as integers (no decimal) — a
float is rejected on the wire.
- Callback mode returns the ack as raw bytes (read via `arrayBuffer()`),
not a typed `SpeakV2AcceptedResponse`; parse `request_id` yourself.
  - **`AgentV1LatencyReport`** — new agent latency-report type.
- **Agent inject-message `interrupt`** — new value on the inject-message
behavior.

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please).

v5.5.0

5.5.0 (2026-06-26)

Features

• regen: flux word timings, diarize_model, profanity_filter (#509) (e8be8ca)

v5.4.0

5.4.0 (2026-06-01)

Adds a pluggable streaming-transport layer (the JS side of the cross-SDK SageMaker work) and fixes the TypeScript types exposed for the WebSocket helper clients. Fully backwards-compatible — every new option is optional with existing defaults, so current callers need no code changes.

Features

• Pluggable transport interface for SageMaker support (#492) (4d43b1b)

  Adds a DeepgramTransport / DeepgramTransportFactory interface (src/transport.ts) and a transportFactory option on DeepgramClient. When set, the listen.v1, listen.v2, speak.v1, and agent.v1 createConnection paths route through a custom transport via a TransportWebSocketAdapter instead of the default WebSocket; REST behaviour is unchanged. This is the seam the @deepgram/sagemaker package plugs into, validated end-to-end against live Nova-3 STT, Flux, and Aura TTS endpoints (including a 400-concurrent-connection burst). Mirrors the same work in the Python and Java SDKs.

• reconnect flag with auto-disable for custom transports (8dec8c9)

  New optional reconnect?: boolean on DeepgramClient (default true, exposed read-only as client.reconnect) controlling wrapper-level retry of streaming connections. Auto-disabled to false when a transportFactory is supplied, so self-retrying transports aren't double-stacked with a second retry layer (which caused storm-on-storm behaviour under burst load). Opt back in with an explicit reconnect: true.

Bug Fixes

• Expose WebSocket helper types (#501) (ac71def) — thanks @asim48-ctrl

  The agent / listen / speak getters were typed against the generated client, which hid the wrapper's createConnection() method and typed Authorization as required. They now expose typed WebSocket wrapper clients, so client.listen.v1.createConnection({ model: "nova-3" }) type-checks and Authorization is optional. Runtime behaviour is unchanged — this is a types-only fix. Fixes #489.

v5.3.0

chore(main): release 5.3.0 (#500)

:robot: I have created a release *beep* *boop*

## 5.3.0 — what's in this release

This release bundles the 2026-05-14 Fern SDK regeneration plus two
follow-up commits on top of #499.

See the full changelog:
[`v5.2.0...v5.3.0`](https://github.com/deepgram/deepgram-js-sdk/compare/v5.2.0...v5.3.0).

### Features

**Diarization v2 batch GA + listen-provider dedup + think/models route
fix — #499**


[`DeepgramEnvironment`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/environments.ts)
reshape (intentional upstream fix, not a runtime break):
-
[`DeepgramEnvironment.Agent`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/environments.ts)
removed;
[`Production`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/environments.ts)
gained an `agentRest` slot.
- The
[`agent.v1.settings.think.models`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/resources/settings/resources/think/resources/models/client/Client.ts)
endpoint now routes via HTTPS instead of WSS.
- Not a breaking change in practice — the old
`DeepgramEnvironment.Agent` would have routed every REST client at
`https://agent.deepgram.com` (404 on `/v1/listen`, etc.) and
`think/models` was misrouted over WSS, so no consumer could have had a
working flow through it. Driven by
[deepgram-docs#886](https://github.com/deepgram/deepgram-docs/pull/886)
+ [#887](https://github.com/deepgram/deepgram-docs/pull/887).

Diarization v2 batch GA (purely additive):
- New optional `diarize_model` field on
[`ListenV1RequestUrl`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/listen/resources/v1/resources/media/client/requests/ListenV1RequestUrl.ts)
and
[`MediaTranscribeRequestOctetStream`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/listen/resources/v1/resources/media/client/requests/MediaTranscribeRequestOctetStream.ts);
new
[`MediaTranscribeRequestDiarizeModel`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/listen/resources/v1/resources/media/types/MediaTranscribeRequestDiarizeModel.ts)
enum.

Listen-provider dedup:
- New top-level exports
[`DeepgramListenProviderV1`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/types/DeepgramListenProviderV1.ts)
/
[`DeepgramListenProviderV2`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/types/DeepgramListenProviderV2.ts).

**Restore `AgentV1SettingsAgentContextListenProviderV2.LanguageHint`
nested namespace
([`ace680a`](https://github.com/deepgram/deepgram-js-sdk/commit/ace680ae192d85dcb5f54268984ccf0866aa62e9))**
- The dedup collapsed the nested
`AgentV1SettingsAgentContextListenProviderV2.LanguageHint` namespace
path. Restored as a one-line alias `type LanguageHint =
DeepgramListenProviderV2.LanguageHint` inside
[`AgentV1SettingsAgentContextListenProvider.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/types/AgentV1SettingsAgentContextListenProvider.ts)
so existing consumers of the nested path keep working.
- Regression coverage in
[`tests/unit/compat-aliases.test.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/tests/unit/compat-aliases.test.ts).

### Bug fixes

**Restore
[`client.fetch()`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/Client.ts)
default `baseUrl` to `api.deepgram.com`
([`56b8ce5`](https://github.com/deepgram/deepgram-js-sdk/commit/56b8ce555c09842e6021fcb889e4aff9e87b151d))**
- After the upstream spec introduced the new `agentRest` slot, the
generator switched the
[`client.fetch()`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/Client.ts)
passthrough helper's default `baseUrl` from `env.base`
(api.deepgram.com) to `env.agentRest` (agent.deepgram.com). Reverted
locally in
[`src/Client.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/Client.ts)
— passthrough should default to the canonical Deepgram REST host. Likely
a Fern generator artifact; this freeze gets dropped once the upstream
fix lands.

### Patches preserved through regen (re-applied in #499)

No prior patches dropped — every existing freeze still addresses surface
the new regen misses.

-
[`package.json`](https://github.com/deepgram/deepgram-js-sdk/blob/main/package.json)
— kept `@types/node ^20.17.57` pin, re-added `@commitlint/cli`,
`@commitlint/config-conventional`, `tsx`, `playwright`, `vite`, `terser`
(generator's `5.2.1` version bump kept).
-
[`src/api/resources/manage/.../keys/client/Client.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/manage/resources/v1/resources/projects/resources/keys/client/Client.ts)
— restored optional `request?` parameter to
`manage.v1.projects.keys.create` (3 spots).
-
[`src/api/types/CreateKeyV1Request.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/types/CreateKeyV1Request.ts)
— restored legacy `CreateKeyV1RequestOne` alias.
-
[`src/api/resources/agent/resources/v1/types/AgentV1Settings.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/types/AgentV1Settings.ts)
— restored interface form of `Agent` with named sub-types +
`AgentReference = Agent | string` opt-in alias.
-
[`src/api/resources/agent/resources/v1/types/AgentV1SettingsAgentListenProvider.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/types/AgentV1SettingsAgentListenProvider.ts)
— restored one-line alias to
[`AgentV1SettingsAgentContextListenProvider`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/types/AgentV1SettingsAgentContextListenProvider.ts).

### New freezes added this cycle (in #499)

-
[`biome.json`](https://github.com/deepgram/deepgram-js-sdk/blob/main/biome.json)
— generator emits `"vcs": { "enabled": false }`, which caused biome lint
to scan the gitignored `examples/browser/deepgram.js` bundle. Restored
`useIgnoreFile: true` and added to `.fernignore` so this patch survives
the next regen.
-
[`AgentV1SettingsAgentContextListenProvider.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/api/resources/agent/resources/v1/types/AgentV1SettingsAgentContextListenProvider.ts)
— see the `LanguageHint` follow-up commit above.
-
[`src/Client.ts`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/Client.ts)
— see the `client.fetch()` bug-fix commit above.

### Validation
- `make build` — passes (CJS + ESM + ESM validation).
- `make test` — passes (535 unit tests + 132 wire tests, including 2 new
regression cases for the `LanguageHint` shim).
- `make test-esm` — passes (5 tests).
- `make lint` — passes (warnings only, no errors).
- `make examples` — 30/36 pass against the real API; remaining 6
failures are env-config (missing `DEEPGRAM_PROJECT_ID`) or transient
network, not regen-related.
- Example 9 (Voice Agent) exercises the rerouted
`agent.v1.settings.think.models.list()` endpoint end-to-end.

### Cross-SDK reference
The `agentRest` env fix has equivalents in
deepgram/deepgram-python-sdk#715 and deepgram/deepgram-java-sdk#59. JS
is the only SDK that additionally needed the
[`client.fetch()`](https://github.com/deepgram/deepgram-js-sdk/blob/main/src/Client.ts)
passthrough patch — Python's
[`client.py`](https://github.com/deepgram/deepgram-python-sdk/blob/main/src/deepgram/client.py)
and Java's `DeepgramClient`/`DeepgramClientBuilder` have no equivalent
passthrough.

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

v5.2.0

5.2.0 (2026-05-12)

Features

• alias AgentV1SettingsAgentListenProvider to *AgentContextListenProvider (150e663)
• preserve AgentV1Settings.Agent sub-types after regen (2efab2d)
• preserve SDK compatibility after regen (d2b8d62)
• restore Agent interface, add AgentReference for string-id flow (4c72d31)
• sdk regeneration 2026-04-30 (#491) (a618282)
• sdk regeneration 2026-05-06 with backward-compat preserved (#497) (2aed53e)

v5.1.0

5.1.0 (2026-04-27)

Features

• sdk regeneration 2026-04-24 (4466be0)

Bug Fixes

• ci: pin JS tooling used by PR title and example checks (7a1c1ea)

v5.0.0

5.0.0 (2026-03-04)

⚠ BREAKING CHANGES

• release new codegen architecture and interfaces (#447)
• the interfaces for the voice agent have changed.
• resolve lint issue
• complete rearchitecture of the Node/JavaScript SDK to work in browsers and Node.

Release

• JS SDK V3 (#208) (2734f69)

Features

• abstract client rebuilt for namespace options (486fd86)
• add a method to send KeepAlive ws messages (#150) (880e943)
• add lint and build to workflow (1710c46)
• add linting for yaml files (f10d10d)
• add markdown linting (899b0e9)
• add missing feature toggles from q1 feature audit (#237) (7f972d1)
• add new models endpoints to SDK (630387d)
• add SpeakClient and example (a3112bc)
• add SpeakLiveClient and LiveTTSEvents (39a535d)
• add speechstarted event to sdk (#235) (1f80bab)
• add support for new beta release of summarisation (#143) (dfecdf2)
• add timeout mechanism to WebSocket mock waitForConnection method (a663956)
• add UtteranceEnd event to sdk (#234) (82f8b2f)
• add warning metadata and summarisation result to response type (#152) (1288778)
• adds support for agent mip_opt_out (#421) (76f9653)
• adds support for IUM (aa6cee2)
• adds support for speak fallback (#414) (8e59b55)
• adds ttl_seconds support (#420) (5ed8004)
• apply luke's suggestions (517ea3c)
• auth: add support for DEEPGRAM_ACCESS_TOKEN as per the other SDKs (7924959)
• Better VTT and SRT (#141) (c3cfc28)
• configurable websocket url (73e53e3)
• convert all legacy options to namespaced options (154ba30)
• errors: enhance WebSocket error output and abstract connection setup (36778e9)
• filler words release (74a4e56)
• filler words release (11b90b8)
• finish it up! (73e53e3)
• finishing touches TTS live client (5da606d)
• i feel like i'm getting nowhere here (73e53e3)
• implement no_delay (726f676)
• implement multi-lingual (034d5b0)
• implements finalize on the liveclient (cee0704)
• improve experience around usage of custom API endpoints (#230) (b779348)
• latest spec fixes (1a37177)
• lint the examples (744fe1a)
• make provider objects generic (b0fc5cb)
• merge in new websocket transmitter changes (565ae2c)
• Merge pull request #238 from deepgram/release/jan-minor-1 (f598162)
• migrate to agent v1 (3a8493b)
• new metadata models endpoints (630387d)
• officially enable aura-2 in live text-to-speech (6b122f9)
• plug live client into new options (8510726)
• plug rest clients into new options (e2f3d8f)
• post-EA changes (630387d)
• rebuilding config and class interfaces (501d6d0)
• release new codegen architecture and interfaces (#447) (4c56fd9)
• release voice agent (73e53e3)
• release voice agent v1 (96753f9)
• Release/3.x.x 20240427 (#271) (7a76f75)
• remove container from websocket speak options (2a03f9a)
• remove specific error typing from the method definitions (#142) (8fbbee7)
• rename reset method to clear, respond with "Clear" payload (b409f8d)
• send back raw buffer instead of ArrayBuffer (298df5c)
• set up linting for typescript (6590d73)
• simplify and standardize examples (81689db)
• sits for JS sdk (#233) (d8de666)
• speak endpoint added (279ad76)
• speak endpoint added (8948856)
• support context length option ([7b22b07](7b22b07c802ff7ee5454eed757...

v5.0.0-beta.4

v5.0.0-beta.4

🎯 What's New

This beta release includes important documentation improvements, enhanced TypeScript support, and better test coverage.

✨ Enhancements

• Enhanced TypeScript Support: Types can now be imported directly (e.g., import { ListenV1Response } from "@deepgram/sdk") in addition to the namespace import, improving IDE autocomplete and discoverability
• Improved Documentation:
  • Clarified that proxy servers are required for browser REST API calls due to CORS header restrictions
  • Added detailed explanation of why the proxy is needed (custom headers not whitelisted)
  • Enhanced migration guide with examples of both direct and namespace type imports
  • Improved README with better examples and explanations
• Better Binary Handling: Fixed critical issue where TTS audio streams were incorrectly parsed as JSON by removing autogenerated message handler
• Improved Examples: All middleware examples converted from JavaScript to TypeScript for better type safety and developer experience
• Enhanced Testing: Added comprehensive tests for custom providers, fallback configurations, and base URL priority handling
• Better Configuration: Improved base URL handling and configuration priority

🧪 Testing Improvements

• Added tests for agent custom providers and fallbacks
• Added tests for base URL configuration priority
• Added tests for custom base URL handling
• Expanded example coverage for middleware use cases

📝 Documentation

• Updated fernignore to protect custom test files
• Improved middleware example documentation

🔧 Internal Changes

• Multiple SDK regenerations to keep up with API changes
• Improved build and test infrastructure

📦 Installation

npm install @deepgram/sdk@5.0.0-beta.4

🔗 Full Changelog

Commits since v5.0.0-beta.3:

• f155120 SDK regeneration
• fbd867c chore: update fernignore
• 16cf0fc Revert "feat: add custom middleware package (TODO: ws support, rest works)"
• 0600353 feat: add custom middleware package (TODO: ws support, rest works)
• 98248d8 feat: test custom providers and fallbacks maybe?
• b3df50c chore: sdk regeneration
• 4e7a1fd feat: convert middleware examples to typescript
• 5af2bab test: add tests for baseurl config priority
• e8f05eb chore: sdk regeneration
• 3fdcf28 SDK regeneration

⚠️ Beta Release Notice

This is a beta release of v5.0.0. Whilst we're confident in the stability, please report any issues you encounter!

• Report issues: https://github.com/deepgram/deepgram-js-sdk/issues
• Docs: https://developers.deepgram.com/
• Discord: https://discord.gg/deepgram