300+ Tools CoveredSource Data Updated Weeklydates

Tool intelligence profile

OpenAI Agents SDK

OpenAI's own agent framework — agents, handoffs, guardrails and sessions over the Responses and Chat Completions APIs, with tracing built in and Python and TypeScript implementations that track each other.

Visit Site →
Type
Agent Framework
Category
Pricing
Free (open source)
Deployment
Self-hosted
Best for
Teams standardised on OpenAI who want multi-agent structure
Last updatedSeptember 20, 2026Open Source

Editor's Take

Best for teams already standardised on OpenAI who want multi-agent structure without adopting a framework with its own worldview — handoffs, guardrails, sandbox agents and first-party realtime voice, all MIT-licensed with no paid tier. The trade-off is direction of travel: the defaults and the built-in tracing assume OpenAI's platform, and other providers work against the grain. Not a fit when you need a durable state machine with cycles and checkpoints, or when provider independence is a hard requirement.

— Egor Burlakov, Editor

Evaluate OpenAI Agents SDK

Comparisons

OpenAI Agents SDK: product and architecture

This OpenAI Agents SDK review covers the framework OpenAI shipped as the production successor to Swarm, its earlier agent experiment. The pitch is a small set of primitives — agents, handoffs, guardrails, sessions — rather than a large abstraction, with tracing built in and Python and TypeScript implementations that track each other. It is MIT-licensed and free; the bill is the model calls. We evaluated it against its documentation, its GitHub repository, and the adoption signals we collect weekly.

Overview

The Agents SDK is MIT-licensed and published by OpenAI. Our 14 September 2026 snapshot recorded about 29,500 GitHub stars on the Python repository, roughly 3.0 million PyPI downloads a week, and about 1.6 million weekly npm downloads of @openai/agents. The TypeScript port is a separate repository maintained by the same team.

Its market position is the default for teams already committed to OpenAI. The SDK's own framing is two design principles: enough features to be worth using, few enough primitives to learn quickly. In practice that means it does not introduce a graph, a chain, or a role system — you orchestrate with ordinary language constructs and the SDK supplies the loop.

The audience is application teams building on OpenAI's models who want handoffs and guardrails without adopting a framework with its own worldview. It is not designed for provider independence, and that is the trade-off to understand before adopting it.

Key Features and Architecture

The primitive set is small and worth listing exactly, because the whole design argument rests on it. Agents are models equipped with instructions and tools, with a built-in loop that runs until the task completes. Handoffs, also usable as agents-as-tools, let one agent delegate to a specialist. Guardrails run input and output validation in parallel with execution and fail fast when a check does not pass. Sessions are a persistent memory layer that maintains working context across turns of one agent loop.

Beyond the primitives, several capabilities matter for production. Sandbox agents run specialists inside isolated workspaces with manifest-defined files, selectable sandbox clients, and resumable sessions — which is the difference between an agent that can write code and one you would let near a repository. Realtime agents build voice applications on gpt-realtime-2.1 with automatic interruption detection and context management. Voice agents assemble a speech-to-text, agent, text-to-speech pipeline for teams not using the realtime model.

Function tools turn any Python callable into a tool, deriving the schema automatically and validating arguments with Pydantic. MCP server tool calling exposes remote Model Context Protocol tools alongside local functions, which is the mechanism for reaching tooling the SDK itself knows nothing about. Human-in-the-loop hooks pause a run for approval.

Tracing is built in rather than bolted on, and it feeds OpenAI's own evaluation and fine-tuning tooling. That is genuinely convenient and also the tightest coupling in the product: the default destination is OpenAI's platform.

Ideal Use Cases

Four situations favour this SDK. The first, and clearest, is a team already standardised on OpenAI models who wants multi-agent structure without a framework migration later. Handoffs are a better-shaped primitive for "route this to the specialist" than a graph edge, and the learning curve is genuinely short.

The second is voice. Realtime agents and the voice pipeline are first-party here, with interruption detection and context management handled, and assembling the equivalent from parts in another framework is real work.

The third is agentic coding or file manipulation, where sandbox agents give you an isolated workspace with resumable sessions rather than an unbounded shell.

The fourth is a team that wants tracing on day one and has no existing observability estate to integrate with. The built-in tracing needs no configuration and connects to evaluation and fine-tuning.

Do not choose the Agents SDK if provider independence is a requirement. It is OpenAI's SDK, its defaults assume OpenAI's platform, and while it can call other models, you are swimming against the design. LiteLLM or a framework like PydanticAI or LangGraph is the better base in that case. It is also the wrong pick if your workflow is a long-running stateful process with cycles and checkpoints — handoffs express delegation, not a durable state machine — and if you need governance features like per-team budgets, those live in a gateway rather than here.

Pricing and Licensing

The SDK is free under the MIT licence, with no paid tier, no seat charge, and no enterprise edition. MIT is permissive enough for commercial use and redistribution, and it satisfies an open-source-only licence requirement for this layer of a stack.

The cost is entirely model spend, billed by OpenAI at the published Responses and Chat Completions API rates. That matters more here than with most frameworks, because several of the SDK's better features increase call volume: guardrails run validation calls in parallel with execution, handoffs mean a task can touch two or three agents, and the built-in loop continues until the task completes rather than after a fixed number of turns. None of that is waste, but it is not free either, and a cost ceiling is worth setting before a loop runs unattended.

The realtime voice features bill against the realtime model, which is priced differently from text and is the line item most likely to surprise. Built-in tracing is part of the OpenAI platform relationship rather than a separate purchase.

Compared with the alternatives, the licensing position is the same as PydanticAI's and LangGraph's — free, MIT or equivalent, no paid tier — and simpler than CrewAI's or LangChain's Freemium model, where an open framework sits beside a paid platform. The real commercial question is not the framework's price but the provider lock-in: a gateway such as LiteLLM, which is free to self-host, is the usual answer for keeping the model bill competitive.

Strengths & Trade-offs

Pros

  • A genuinely small primitive set — agents, handoffs, guardrails, sessions — that a team can learn in an afternoon and still use in production.
  • MIT-licensed with no paid tier, so there is no relicensing risk and no feature behind a plan.
  • Sandbox agents with resumable sessions and manifest-defined files, which is the safe shape for code-writing agents.
  • First-party realtime voice and a voice pipeline, both of which are significant work to assemble elsewhere.
  • Python and TypeScript implementations maintained by the same team, so a split frontend and backend team can share a mental model.

Cons

  • Designed around OpenAI's platform. Other providers are reachable but are not the happy path, and the defaults keep pulling back.
  • Built-in tracing defaults to OpenAI's own destination, so an existing observability estate is an integration rather than a given.
  • Handoffs are delegation, not durable state. A long-running workflow with checkpoints and cycles is not what this expresses.
  • No governance layer — budgets, per-team keys, rate limits — so a multi-team deployment needs a gateway in front of it.
  • Fast-moving: the Python package was at v0.22.3 in mid-September 2026, and a sub-1.0 version number is a real signal about API stability.

Alternatives to OpenAI Agents SDK

The reviewed substitutes for OpenAI Agents SDK among the agent frameworks, and what would make each one the better answer.

Direct alternatives

Reviewed substitutes: products bought for the same job, where a team picks one.

LangGraph
Both are adopted as the primary agent framework for an application. Handoffs and graphs are two ways to express the same orchestration decision.
Google Agent Development Kit
The two large model vendors' own agent frameworks. A team building agents adopts one of them or a vendor-neutral framework; it does not adopt both.
PydanticAI
Both are MIT-licensed Python agent frameworks with a small primitive set. The decision between them is provider independence against first-party OpenAI features.

Other approaches

A different approach to the same problem. Each substitutes only for the workload named beside it.

CrewAI
Both appear on the same shortlist and answer different shapes. CrewAI is chosen for role-based crews a product team can read; the OpenAI Agents SDK is chosen as the simplest path to production, with agents, handoffs and tracing and less abstraction to learn. Conditional: CrewAI carries no provider commitment, and the SDK carries no role-and-task model, so neither is a drop-in for the other's reason for existing.Applies to: Coordinating several agents on one task. CrewAI stands in when the division of labour is the design and must be legible; the OpenAI Agents SDK stands in when the team is on OpenAI and wants handoffs and tracing without a new paradigm.
Agno
Same fork, different vendor: the OpenAI Agents SDK is the shortest path to production for a team committed to OpenAI models, while Agno is model-agnostic with its own memory and knowledge layer and no provider commitment. Reviewers describe Agno as the simpler development experience for common RAG use cases, which is the same claim the SDK makes for agent loops -- so the choice is which lock-in a team accepts, not which is more capable.Applies to: A Python agent with sessions, memory and a knowledge base. The OpenAI Agents SDK stands in when the models are OpenAI's and hosted retrieval is acceptable; Agno stands in when the team wants to stay model-agnostic and hold its own vector store.
LangChain
Both are on the 2026 shortlist and the guidance separates them on what a team is optimising: the OpenAI Agents SDK is described as the simplest path to production, LangChain as the most flexible with the broadest tool ecosystem and the steepest learning curve. Conditional because the SDK's simplicity is bought with provider commitment -- it is OpenAI's own SDK, and a team that needs to swap models freely is choosing LangChain for exactly that.Applies to: Getting an agent into production. The OpenAI Agents SDK stands in for LangChain when the models are OpenAI's and shipping quickly matters most; LangChain stands in when the team needs provider portability or the wider tool ecosystem.
Semantic Kernel
Two vendor-aligned frameworks separated by the estate rather than the feature list: Semantic Kernel for .NET and Azure with enterprise process orchestration, the OpenAI Agents SDK for teams committed to OpenAI models wanting the shortest path to production. Semantic Kernel is superseded: Microsoft merged it and AutoGen into the Microsoft Agent Framework, GA April 2026, and SK v1.x now receives critical bug and security fixes only, with new features built in the successor. It stays a real option because a .NET estate running it today is supported into 2027, and the catalogue records the supersession on its own page -- but the pair is a choice between a live framework and one on a known sunset, and saying so is the point.Applies to: Picking an agent framework that matches the stack already in place. Semantic Kernel stands in for a .NET or Azure shop already running it; the OpenAI Agents SDK stands in for a Python team on OpenAI models. A fresh Microsoft build belongs on the Agent Framework successor.
Dify
The same cross-class trade one layer over: Dify's visual canvas with branching, loops and a built-in knowledge base delivers the agent-and-RAG application without SDK code, which is the whole of what a team reaches the OpenAI Agents SDK for on a straightforward build. Conditional: the SDK gives handoffs, tracing and programmatic control Dify does not expose, and Dify gives an authoring surface a non-engineer can use.Applies to: Delivering an internal agent or RAG chatbot quickly. Dify stands in for the OpenAI Agents SDK when engineering time is the constraint and the flow fits a canvas; the SDK stands in when the agent needs handoffs, tracing or logic beyond what the builder exposes.
LlamaIndex
OpenAI's File Search tool is described as RAG functionality with less work: it parses and chunks documents, creates and stores embeddings, and runs vector plus keyword search with query optimisation and reranking built in, billed at $2.50 per thousand queries and $0.10 per GB per day. Paired with the Agents SDK that covers the agent loop, handoffs and tracing, it removes the reason an OpenAI-committed team would build an ingestion and index layer at all. Conditional: it is a single hosted store with no connector ecosystem, and the substitution ends the moment a second model provider or a self-hosted index is required.Applies to: An agent over a modest document set on OpenAI models only. The Agents SDK with hosted File Search stands in for LlamaIndex when the corpus is standard and hosted retrieval is acceptable; LlamaIndex stands in when the team needs its own index, other providers, or connectors beyond file upload.
Explore all OpenAI Agents SDK alternatives →

Public signals

About these signals

Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.

703 GitHub commits 90d29.6k GitHub stars0 vulnerabilities across 2 packages

See all signals from 5 sources
Source
Signals
Last updated
GitHub
Commits 90d:703Stars:29.6k
September 21, 2026
PyPI
Weekly downloads:3.0M
September 21, 2026
npm
Weekly downloads:1.3M
September 21, 2026
Product Hunt
Comments:4Rating:5.0/5Reviews:848Votes:121
September 21, 2026
OSV
Package vulnerabilities:0 vulnerabilitiesacross 2 packages

npm · @openai/agents@0.18.0 · PyPI · openai-agents@0.22.3

September 21, 2026

Discussed on Hacker News

Recent Hacker News threads mentioning OpenAI Agents SDK.

Frequently asked questions

Is the OpenAI Agents SDK free?

The SDK itself is free and MIT-licensed, with no paid tier and no seat charge. You pay OpenAI for the model calls it makes, at the published Responses and Chat Completions API rates.

Can the Agents SDK call models other than OpenAI's?

It can, but it is not the happy path. The defaults, the built-in tracing destination, and the realtime voice features all assume OpenAI's platform. If provider independence is a requirement, PydanticAI or LangGraph behind a gateway such as LiteLLM is a better base.

How is this different from Swarm?

The Agents SDK is the production successor to Swarm, which OpenAI published as an experiment. The primitives are recognisably the same idea, with handoffs, guardrails, sessions, sandbox agents, and tracing added for production use.

Does it handle long-running stateful workflows?

Not really. Handoffs express delegation between agents and sessions hold working context within a run. A process with cycles, conditional branches, and checkpoints that survive a restart is what LangGraph is for.

Related Agent Frameworks

Other agent frameworks in the catalog. Same kind of product, not a substitution recommendation.