openai/openai-agents-js

A lightweight, powerful framework for multi-agent workflows and voice agents

What it solves

This SDK provides a lightweight framework for building multi-agent workflows in JavaScript and TypeScript. It simplifies the orchestration of LLMs that can take actions, maintain state, and collaborate through handoffs, reducing the complexity of managing conversation history and agent interactions.

How it works

The framework is built around several core concepts:

  • Agents: LLMs configured with specific instructions, tools, and guardrails.
  • Sandbox Agents: Specialized agents that operate within a filesystem workspace and sandbox environment for longer-running tasks.
  • Realtime Agents: Low-latency voice agents designed for spoken interactions in the browser via WebRTC.
  • Handoffs: A mechanism allowing agents to delegate tasks to other agents, enabling multi-agent collaboration.
  • Tools: Integration with functions, MCP, and hosted tools to allow agents to take real-world actions.
  • Guardrails: Configurable safety checks for validating input and output.
  • Sessions: Automatic management of conversation history across agent runs.
  • Tracing: Built-in tracking for debugging and optimizing agent workflows.

Who it’s for

Developers using Node.js, Deno, or Bun who want to build sophisticated AI agents that can interact with filesystems, handle voice interactions, or work in teams of specialized agents.

Highlights

  • Provider-agnostic: Supports OpenAI APIs and other providers.
  • Sandbox environments: Built-in support for local (macOS/Linux) and Docker-based sandboxes for executing commands and managing files.
  • Human-in-the-loop: Built-in mechanisms for involving humans in the agent's execution flow.
  • Realtime voice support: Native integration for low-latency spoken interactions.