
OpenAI Agents JS
https://github.com/openai/openai-agents-js- Category
- AI Agents
- Rank
- No. 290Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- openai
- GitHub
- 3.8k stars
- Latest release
- v0.17.2
- Date
About
Lightweight, powerful framework for multi-agent workflows and voice agents in JavaScript/TypeScript — built by OpenAI for production.
What it does
It supplies building blocks for model-driven applications that can call tools, pass work between specialized agents, validate inputs and outputs, retain conversation history, stream results, and record traces. Separate modes handle ordinary text tasks, filesystem-backed work, and low-latency browser voice sessions.
Why it's ranked here
The SDK covers unusually broad orchestration needs within one TypeScript library: handoffs, MCP tools, guardrails, human approval, sessions, tracing, sandbox work, and realtime audio. Provider abstraction and scripted testing utilities strengthen the foundation. Beta sandbox support and demanding runtime requirements keep the verdict measured.
What's good
The design exposes concrete controls instead of hiding orchestration behind prompts. Developers can set tool guardrails, retry policies, error handlers, approval steps, session behavior, and tracing processors. Realtime support spans WebRTC, WebSocket, and SIP, while scripted models and sandbox sessions make agent behavior testable without relying entirely on live model calls.
Tradeoffs
Node users need version 22 or later. Cloudflare Workers support remains experimental, and sandbox agents remain beta. The local Unix sandbox client supports macOS and Linux, while Windows requires Docker or a hosted client. Browser voice applications also need a server to mint short-lived client tokens securely.
How to use it well
Choose it for TypeScript products that need coordinated model calls, tool execution, safety checks, persistent conversations, or spoken interaction. Start with a text agent, add handoffs only where specialization helps, and use scripted test doubles before live integration tests. It does not provide the sandbox environment itself; select a local, Docker, or hosted client.
Technical notes+
packages/agents/src/index.ts installs OpenAIProvider as the default model provider, disables response caching for WebSocket models, configures the OpenAI tracing exporter, and re-exports core, OpenAI, and realtime surfaces. packages/agents-core/src/index.ts exposes runners, sessions, MCP transports, guardrails, tools, retries, tracing, errors, and protocol types, then registers defaultProcessor() at module load. packages/agents-openai/src/index.ts adds Responses, Chat Completions, hosted tools, conversation sessions, compaction, and tracing export. packages/agents-realtime/src/index.ts exposes WebRTC, WebSocket, SIP, audio utilities, guardrails, and realtime session primitives. packages/agents-core/src/testing/index.ts provides scripted model and sandbox fixtures. Root package.json defines a private pnpm workspace with Vitest unit, integration, coverage, example, and released API contract tasks.
Observed
- Primary language
- JavaScript and TypeScript
- Install surface
- npm package @openai/agents with zod
- Interface
- Library SDK with text, sandbox, realtime, MCP, and hosted-tool surfaces
- Supported runtimes
- Node.js 22 or later, Deno, and Bun
- Experimental platform
- Cloudflare Workers with nodejs_compat enabled
- Testing structure
- Vitest unit, integration, coverage, example, and released API contract scripts are defined
- Package structure
- Private pnpm workspace with separate core, OpenAI provider, realtime, and extensions packages
Read from README.md, package.json, packages/agents/src/index.ts, packages/agents-core/src/index.ts, packages/agents-openai/src/index.ts, packages/agents-realtime/src/index.ts, packages/agents-extensions/src/index.ts, packages/agents/src/utils/index.ts, packages/agents/src/sandbox/index.ts, packages/agents/src/testing/index.ts, packages/agents/src/realtime/index.ts, packages/agents-core/src/types/index.ts, packages/agents-core/src/utils/index.ts, packages/agents-core/src/sandbox/index.ts, packages/agents-core/src/testing/index.ts.
What it can do
Create multi-agent workflows
JavaScript/TypeScript code and agent configurations → Coordinated multi-agent system
Build voice-enabled agents
Voice input and agent logic → Interactive voice agent
Execute agent orchestration
Multiple agent definitions and workflow rules → Automated agent interactions and results
Process natural language conversations
Text or voice messages → Contextual agent responses
Integrate OpenAI models into workflows
API configurations and model parameters → AI-powered agent behaviors
Handle real-time voice interactions
Live voice streams → Real-time voice responses
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.