Migration path
A governed home for the workflows you built in Agent Builder.
OpenAI notified developers on June 3, 2026 that Agent Builder is deprecated, and it is scheduled to shut down on November 30, 2026. Your workflow architecture does not have to retire with the canvas you drew it on.
Sources: OpenAI deprecations and Migrate from Agent Builder, both retrieved 2026-08-03.
What is changing, precisely
Agent Builder — the visual workflow canvas — is scheduled to shut down on November 30, 2026. On the same date the Evals dashboard and API are scheduled to shut down; existing evals become read-only on October 31, 2026.
This is not OpenAI stepping back from agents. Their own guidance points Agent Builder users to the Agents SDK or ChatGPT Workspace Agents, and ChatKit remains available. What is going away is the visual design surface — the place where the shape of the workflow lived.
OpenAI is candid about what a migration carries over:
This process does not convert your workflow graph or guarantee that every behavior transfers unchanged.
OpenAI, Migrate from Agent Builder
What Everform does with your export
Export your workflow as code from Agent Builder — Code in the top navigation, then Agents SDK, then TypeScript or Python. That is OpenAI’s documented export, and it is the file Everform reads.
Everform reads the declared agents — names, instructions, models, tools, guardrails, handoffs — into a workflow on a real canvas, scores its architecture with a deterministic engine, and puts it somewhere governed: oversight visible per step, evidence labelled by how it was obtained, and a record of what changed.
Everform is not a replacement for running your agents. Keep running them wherever you choose. This is the layer above: the architecture, and the evidence about it.
What we won’t pretend
Everything Everform reads from your export is labelled declared_by_source — a description of what your export declares, never proof that it works. An architecture score on an imported workflow is an estimate until real runs produce evidence.
Code outside the declared agents — your control flow, your tool implementations — is recorded as unmapped rather than guessed at. The import reports what it could not interpret alongside the workflow, instead of quietly leaving it out.
Automatic import is not open yet. We are migrating the first workflows hand-in-hand with the people who built them, so the reader is proven against real exports before it ever runs unattended on yours.
Export from Agent Builder
Open your workflow, choose Code, then Agents SDK, then TypeScript or Python, and copy the complete export. Worth doing well before the scheduled date rather than close to it.
Bring it to us
We migrate the first workflows together with their owners rather than handing you an importer we have not yet proven on a real export. You keep the original either way.
See your architecture, honestly
Your agents and handoffs land on the canvas with an architecture score, and the import comes with a record of what it read, what it could not, and where all of it came from.
Bring your workflow over
Free to start. The first migrations are done with you directly — no importer running unattended on a workflow nobody has checked.