Automation

OpenAI Agent Builder: the November 2026 shutdown and what to use instead

OpenAI Agent Builder is scheduled to shut down on 30 November 2026. Here is what it is, the routes OpenAI suggests, when rebuilding the agent in n8n makes more sense and what to do before that date.

What OpenAI Agent Builder is and when it shuts down

If your team built an agent in OpenAI Agent Builder, it has to move before the shutdown. OpenAI gives the date on its deprecations page, after notifying developers in early June. Existing users can keep using it during the transition window.

Agent Builder is a visual canvas for building multi-step agent workflows. You start from a template or a blank canvas, add and connect nodes, define typed inputs and outputs, and test the workflow with live data in a preview that shows each node running. There were two ways to deploy: embed it with ChatKit by passing the workflow ID, or download its Agents SDK code and run it on your own infrastructure.

OpenAI launched it in October 2025 as part of AgentKit, a set that bundled Agent Builder, a Connector Registry for admins and ChatKit, the toolkit for embedding an agent chat in a website or app. ChatKit stays available. OpenAI's Evals platform, used to evaluate agents, shuts down on the same day and has been read-only since 31 October 2026.

What Agent Builder cost, and what the alternatives cost

The Agent Builder guide lists no separate fee for the canvas. Agents run on OpenAI's API, whose pricing page bills each model's tokens and charges some tools on top. In November 2026, web search costs $10 per 1,000 calls plus search content tokens, file search $2.50 per 1,000 calls, and storage for files uploaded to ChatKit $0.10 per GB per day, with 1 GB free per account each month.

With the Agents SDK you still pay for API usage, plus the server that runs your code and the hours of whoever maintains it. Workspace Agents run inside a ChatGPT Business, Enterprise or Edu workspace, and their usage comes out of that workspace's allowance and credits, at the rates OpenAI publishes in its help centre.

In n8n, the cost splits between the plan, or the server if you host it, and the usage of the model you choose, billed by its provider. In all three cases, add the time spent building and maintaining the agent.

Four routes for an Agent Builder workflow

Decide workflow by workflow. One company can move one agent to code and another to n8n.

  • Agents SDK, if the workflow should stay in code

    This is OpenAI's recommendation for workflows that should continue as code. You export it in Python or TypeScript, add it to your application and deploy it on your own infrastructure. It fits if you have developers and the agent is part of a product.

  • Workspace Agents, if instructions are enough

    OpenAI suggests them for use cases that work well with natural-language prompting. They need a ChatGPT Business, Enterprise or Edu workspace. OpenAI warns that workflows with strong determinism at their core may not migrate faithfully.

  • n8n, if the agent lives among your tools

    If the workflow reads email, updates the CRM and writes to a spreadsheet, n8n has ready-made integrations and an AI agent node that works with OpenAI models and those of other providers. You rebuild the graph by hand, using the export as a reference, and you can host it on your own server.

  • None, if nobody uses it

    Ask who really uses it. An agent nobody has opened in months does not deserve a migration. Archive the export and the instructions in case you ever need it back.

What to do before 30 November 2026

This list works even if you have not chosen a route yet.

  1. 01

    Take stock

    List each Agent Builder workflow, who uses it, which data it touches and where it is deployed: in ChatKit on your website, in your own code or only in testing.

  2. 02

    Export the code

    Open each workflow, select Code in the top bar, choose Agents SDK, pick Python or TypeScript and copy the full export. Keep it alongside each agent's instructions.

  3. 03

    Check your ChatKit widgets

    OpenAI's hosted ChatKit integration uses the Agent Builder workflow ID. OpenAI does not say what happens to those widgets at shutdown and recommends the advanced integration, on your own server, for new or migrated apps. Plan as if the current widget will stop answering.

  4. 04

    Save test cases

    Collect around twenty real questions with the right answer for each. They will show whether the migrated agent behaves like the original. If you used Evals, export what you need while it is read-only.

  5. 05

    Run the new version in parallel

    Build the agent in its new home, run the test cases through it and compare the answers with the original workflow while it is still live.

Example

A Cardiff online shop moves its returns agent

In this illustrative case, an online clothing shop in Cardiff with four staff built an agent in Agent Builder that answers returns questions, embedded on its website with ChatKit. It reads the returns policy and, when a customer gives an order number, checks the order status in the shop.

Taking stock, the team sees that the agent depends on three systems: the shop, the customer service inbox and the sheet where returns are logged. Nobody codes, so the Agents SDK is out. Workspace Agents live in the team's ChatGPT workspace, while this agent talks to customers on the website.

They rebuild it in n8n. They export the code to keep the instructions, then build a workflow with an AI agent node, a tool that looks up orders and another that logs the return in the sheet, connected to a chat on their website. Late returns always go to a person.

Before switching, they run the twenty test cases through both agents and compare the answers. Afterwards they track each week how many conversations close without help, how many go to a person and which errors appear, by type.

Choosing your agent's new home

If your agent is part of a product maintained by developers, the Agents SDK is the shortest route. If it lives inside ChatGPT and your team already works there, try Workspace Agents. If it connects several company tools and you want to be free to change models, n8n gives you more room.

Obinova is an AI agency based in Barcelona and Paris. Through our n8n automation service, we rebuild AI workflows and agents with your team, connected to your email, CRM and internal tools, and document them so your people can keep improving them. Show us your Agent Builder workflow and we will tell you whether it belongs in n8n or on another route.

Frequently asked questions

When does OpenAI Agent Builder shut down?

According to OpenAI's deprecations page, the shutdown is scheduled for 30 November 2026. OpenAI notified developers on 3 June 2026, and existing users can keep using it during the transition. OpenAI does not say what happens to workflows after that date, so export them beforehand.

What is OpenAI AgentKit?

It is the name OpenAI gave in October 2025 to a set of tools for building and deploying agents: Agent Builder as a visual canvas, a Connector Registry for admins and ChatKit for embedding the chat in a website or app. Of those parts, Agent Builder shuts down in November 2026 and ChatKit remains available.

What happens to ChatKit when Agent Builder closes?

ChatKit stays available. What depends on Agent Builder is the hosted integration, which uses an Agent Builder workflow ID. For new or migrated apps, OpenAI recommends the advanced integration, where you run the agent on your own server with the ChatKit Python SDK.

Can I import an Agent Builder workflow into n8n?

There is no direct import. Agent Builder's export produces Agents SDK code, not a file n8n can open. It works as a reference: each agent's instructions, its tools and the order of steps are carried over by hand into an n8n workflow, then tested against the same cases as the original.

n8n or the Agents SDK: which suits a small business?

It depends on who will maintain the agent. The Agents SDK needs developers to write, deploy and monitor the code. n8n lets you build and adjust the workflow visually, connects to many common tools and works with models from several providers. For a small business without a technical team, n8n is usually easier to keep running.

Let's find your agent a new home

Book a free 30-minute video call. We will look at your Agent Builder workflow together and tell you frankly which route fits and what it involves.

Free 30-minute diagnosis →
Free diagnosisWhatsApp

WhatsApp · Reply within 2 hours during business hours