OpenAI Agents SDK

The OpenAI Agents SDK is a Python framework (Python 3.10+) for building multi-agent workflows. It is provider-agnostic: it supports the OpenAI Responses and Chat Completions APIs and, through any-llm and LiteLLM, 100+ other LLMs. A JavaScript/TypeScript version is maintained separately.

An agent is an LLM configured with instructions, tools, guardrails, and handoffs. The SDK offers four ways to run agents: a text Agent for ordinary workflows, a SandboxAgent that works inside a container to inspect files, run commands, apply patches, and keep workspace state across long tasks, a RealtimeAgent for server-side voice and multimodal sessions over WebSocket, and a VoicePipeline that chains speech-to-text, an agent workflow, and text-to-speech. Sandbox clients include a local Unix sandbox on macOS and Linux and a Docker sandbox.

Features

• Agents: LLMs configured with instructions, tools, guardrails, and handoffs
• Sandbox agents: agents preconfigured to work in a container for long-horizon tasks, with manifests, permissions, and workspace entries such as Git repositories
• Realtime and voice agents: voice agents on gpt-realtime-2.1 and voice pipelines combining STT, an agent, and TTS
• Handoffs and agents as tools: delegate specific tasks to other agents
• Tools: function tools, MCP servers, and hosted tools
• Guardrails: configurable checks that validate input and output
• Human in the loop: built-in mechanisms for involving people across agent runs
• Sessions: automatic conversation history across runs, with SQLite, SQLAlchemy, Redis, and encrypted session backends
• Tracing: built-in tracking of agent runs for viewing and debugging workflows