Official OpenAI TypeScript and JavaScript client library that lets you call the OpenAI REST API from Node.js or other JS/TS environments. It provides convenient methods for common APIs such as the Responses API, supports streaming via SSE, and includes helpers for features like file uploads and workload identity authentication.
Project status
- Actively maintained, with an upstream push today and several updates in the past few days.
- Recent updates are arriving frequently, sometimes multiple times per day; this shows the recent pace, not necessarily a long-term cadence.
AI summary generated
Recent updates
v7.53.0
v7.53.0 adds support for delivering agent session errors as stream events. The implementation introduces an internal stream delivery mode while keeping the default legacy behavior, and updates dependencies and development tooling.
Featuresv7.52.0
Version 7.52.0 adds the `gpt-6-cyber` model to the Responses API model options. It also updates the Cloudflare Worker ecosystem test's development dependency `follow-redirects` from 1.16.0 to 1.16.2; the diff shows no breaking changes.
Featuresv7.51.0
v7.51.0 adds three data-residency regions, GPT-6 Rosalind to the model types, and expanded image-generation and audio options. It also adds agent turn error typing and updates a development-only dependency; the diff shows no public method signature removals or documented breaking changes.
Featuresv7.50.0
Version 7.50.0 adds 12-hour prompt-cache retention and cache-write usage details, and introduces project-level service-tier settings for organization administrators. It also updates Puppeteer in the browser ecosystem tests and clarifies generated-file ownership guidance.
Featuresv7.49.0
Version 7.49.0 adds Admin API support for registering and managing organization workload identity providers, plus federation rules that bind verified identity attributes to project service accounts. The diff adds new resources, types, API documentation, and tests, with no changes to existing public APIs evident.
Featuresv7.48.0
v7.48.0 expands the Live Responses delegation tool and choice types, and improves WebSocket previews by preserving sparse output facts instead of discarding them when scaffolding is missing. It also updates development dependencies used by several ecosystem test projects.
BreakingFeaturesv7.47.0
Version 7.47.0 adds typed support for Responses WebSocket warmup. The change introduces an optional `generate` field on standard and beta Responses WebSocket client events, allowing request state to be prepared without generating output.
Featuresv7.46.0
Version 7.46.0 adds API support for new agent, image, organization, SIP, Realtime, and Responses capabilities, plus a Realtime WebSocket bug fix. The diff also changes WebSocket send behavior and removes generated-code policy workflows, changes not clearly described in the release notes.
Featuresv7.45.1
v7.45.1 clarifies prompt cache breakpoint TTL documentation and adds compatibility for google-auth-library v11 in the Vertex provider. The supplied diff also changes repository policy and CI workflows, but shows no public TypeScript API signature or runtime behavior changes.
v7.45.0
v7.45.0 adds the gpt-6.1-terra model identifier to the SDK's Responses and shared model types. The remaining changes are dependency updates and ecosystem-test tooling changes; the diff shows no breaking changes to the SDK's public runtime API.
Featuresv7.44.0
v7.44.0 adds scheduled agent automations and invocation APIs, prompt-cache breakpoint lifetimes, and expanded Live Decision item types. It also documents the shutdown of gpt-image-1 and updates the Responses image-generation default to gpt-image-2. The Live Decision union changes are explicitly called out as a breaking change.
BreakingFeaturesv7.43.0
v7.43.0 adds broader Decisions input support, including array-based question instructions and additional input message variants. The release also updates development dependencies and exposes additional Chat Completions parsing types; the diff shows mostly export and compatibility-preserving refactors beyond those changes.
BreakingFeaturesv7.42.0
Release v7.42.0 primarily adds a new Audio Voices deletion API, returning a typed VoiceDeleted object. It also improves performance characteristics in two areas, avoiding repeated hashing when preparing agent file paths and reducing the cost of redacting very long log values. Finally, it updates documentation around project API key deletion behavior for service account owned keys.
Featuresv7.41.0
v7.41.0 adds a new Vertex AI provider that enables using the standard OpenAI client to call the Responses API via Google Vertex. The release notes only mention this feature addition.
BreakingFeaturesv7.40.0
v7.40.0 adds new API capabilities for voice consents, expanded usage accounting (30-minute cache write counts), and support for compaction trigger IDs. It also includes multiple fixes around request encoding/response parsing and WebSocket behavior, especially defaults for large payload handling and compression.
Featuresv7.39.0
v7.39.0 introduces an opt-in Realtime feature for bounded connection recovery, intended to transparently attempt reconnection and retain unsent client commands for later flushing. The release notes also mention a separate documentation-only chore (neutral moderation examples), but the code diff shows substantial new Realtime reconnection APIs and behavioral rules that are not spelled out in the release notes.
Featuresv7.38.0
v7.38.0 primarily adds typed Live sideband WebSocket event definitions, extending the SDK’s discriminated union for sideband server events. It also updates the beta Agent and Session APIs to accept an additional service tier value, ultrafast.
Featuresv7.37.0
This release (v7.37.0) adds a new external storage status value, 'degraded', and updates realtime translations internals to use generated translation websocket bindings. It also updates documentation for GPT-4 model alias shutdown notices. Overall changes are mostly additive, with a notable internal refactor in realtime translations URL/query preparation.
Featuresv7.36.0
v7.36.0 adds support for a new `safety_identifier` field on image-related requests and surfaces the effective `safety_identifier` on image-related responses/events. Release notes only call out the feature at a high level, but the code introduces additional request serialization and parameter precedence behavior not explicitly documented.
Featuresv7.35.0
v7.35.0 adds new API capabilities for Decisions prompt caching, prompt-based voice creation, enhanced response error diagnostics, and additional Live session functionality for audio fixture playback and Decisions events. It also refactors internal WebSocket connection lifecycle code that underpins several Realtime and Live transports.
Features