Temporal gives you an intuitive way to code workflows that always execute without failure. With Temporal, you spend less time writing plumbing logic and more time writing business logic, to boost productivity by 10x with the same technology that supports applications at Nvidia, Lovable, and Netflix, as well as every Snap story and every Taco Bell order.

Temporal is like React for asynchronous and scheduled tasks. It lets us build complex workflows with confidence and resilience, making sure they work perfectly every time.

Temporal is changing the world. Durable execution is the future.

It powers our multi-step workflows across agent actions. Temporal gives us durability, visibility, and precise control. Battle-tested orchestration with strong developer UX.

Temporal is like React for asynchronous and scheduled tasks. It lets us build complex workflows with confidence and resilience, making sure they work perfectly every time.

Temporal is changing the world. Durable execution is the future.

It powers our multi-step workflows across agent actions. Temporal gives us durability, visibility, and precise control. Battle-tested orchestration with strong developer UX.

Rate-limiting? Brittle tools? Rapid iteration? We’ve got it handled. OpenAI Agents SDK + Temporal means your agents now remember state, bounce back from crashes, and survive real-world chaos without you writing error-handling and orchestration code.

Hi, Product Hunt Community, I’m the Co-Founder and CTO of Temporal and I’m excited to launch this integration with OpenAI’s Agents SDK in Public Preview.

What is it?

This new integration between the OpenAI Agents SDK and Temporal’s Python SDK lets you build production-ready AI agents, fast. What does that mean? It’s easy to prototype agents, but they fall apart in production. Rate-limiting, crashes, flaky APIs and unreliable tools. These issues sound new, but they aren’t. They’re a different flavor of distributed system challenges. Temporal has been solving these challenges for years to make distributed systems more accessible and reliable.

With this integration, you can use abstractions and best practices from OpenAI with Temporal’s orchestration, Durable Execution, and scale to build agents for the real world. Add several lines of Temporal code to agents you already built with OpenAI’s framework or start building with both from the beginning.

It’s available in Public Preview in Temporal’s Python SDK, open-source under the MIT license.

What’s included?

Lightweight, easy-to-use Agents SDK with very few abstractions: the OpenAI Agents SDK is straightforward to use and provides industry best practices.
Durable Execution for agents: Temporal’s crash-proof execution provides automatic state persistence, retries, seamless recovery across failures, long-running task handling, and human-in-the-loop. This approach also preserves tokens since you don't need to rerun LLMs to get back content you lost.
100% code: The Agents SDK & Temporal both provide a plain-code developer experience that is flexible, intuitive, and gives you the freedom to develop how you want.
Built-in observability and debugging tools: includes local replay, support for unit testing, and traceability of every step in an application.
Massive horizontal scale: run thousands (or millions) of agents in parallel, with the same technology that supports applications at Nvidia, Lovable, and Netflix, as well as every Snap story and every Taco Bell order.

We’re excited to see what you build. You can read more about the integration here. Try it out, and let us know what you think!

@maximfateev Could this power multi-agent research pipelines or autonomous support bots

@maximfateev @masump 💯 Absolutely!

The Open AI Agents SDK follows a model of creating smaller, for-purpose agents - I've come to think of them as micro agents (akin to microservices) that are then orchestrated together to delivery bigger values - like micro-agent research or any other manner of complex processes.

Adding Temporal means that as soon as one of these composite agents starts, it will run to completion - even if all sorts of chaos rains down on them!

@maximfateev this integration makes so much sense. The durable execution for agents is exactly what's been missing; I've lost too many tokens to crashes. What's the memory overhead like?