Join Waitlist

OaiAgentBld · Waitlist

Move off OpenAI Agent Builder before Nov 30 — know whether to export to the Agents SDK or a Workspace Agent, and what to retest

  1. Deprecation announced. Existing users can keep using Agent Builder during the transition window.

  2. Agent Builder scheduled to shut down.

  3. ChatKit remains available

    Agent Builder-hosted workflows in ChatKit are supported during the transition window; for migration, use ChatKit's custom server integration.

Same timeline: Evals become read-only Oct 31, 2026 and shut down Nov 30, 2026; reusable prompts (v1/prompts) shut down Nov 30, 2026.

Dates per OpenAI "Deprecations" and "Migrate from Agent Builder". Export doesn't convert your workflow graph or guarantee every behavior transfers unchanged.

OpenAI announced Agent Builder's deprecation on Jun 3, 2026, and it's scheduled to shut down on Nov 30, 2026. ChatKit remains available. Answer 5 quick questions — Agents SDK or ChatGPT Workspace Agent, how many workflows and how critical, whether you rely on connected apps or a ChatKit embed, how deterministic your flows are, and whether you've tried the Code export. Get your path, export steps, a rebuild list, and a verification checklist. Never paste API keys, tokens, workflow IDs, workflows, or code.

Join Waitlist See how the 5-question check works

Independent tool. Not affiliated with or endorsed by OpenAI.

Problem

Key dates: Jun 3 · Nov 30 · ChatKit stays

OpenAI Agent Builder is scheduled to shut down on Nov 30, 2026 — and teams already running workflows in Builder lack a lightweight decision layer that turns "target × count and criticality × connected dependencies × determinism × export progress" into a path choice, export steps, a rebuild list, and a verification checklist.

Before

  • Jumping between five OpenAI docs pages.
  • Thinking ChatKit is shutting down too — or that export means done.
  • Finding out in December that your ChatKit embed's hosted workflow is gone.

After — with OaiAgentBld

  • 5 questions → only the path and steps that apply to you.
  • Agent Builder shuts down Nov 30; ChatKit stays; export doesn't convert the graph.
  • ChatKit moved to your own server + Agents SDK, verified side by side during the transition window.

Official timeline (OpenAI "Deprecations")

DateStatusWhat it means
Jun 3, 2026AnnouncedAgent Builder deprecation announced; existing users can keep using it during the transition window. ChatKit remains available.
Oct 31, 2026EvalsThe Evals platform becomes read-only (graders are part of the transition); Agent Builder's Evaluate feature uses trace graders.
Nov 30, 2026Scheduled shutdownAgent Builder scheduled to shut down; Evals and reusable prompts (v1/prompts) also shut down that day.
Transition windowChatKitChatKit itself stays; integrations backed by Agent Builder-hosted workflows are supported only during the transition window — for migration, use ChatKit's custom server integration.

OpenAI defines shut down as no longer accessible. OpenAI doesn't spell out what happens to published workflows (via ChatKit or the API) after Nov 30, 2026 — so plan to finish export and migration before then.

Solution

What OaiAgentBld does

An Agent Builder → Agents SDK / Workspace Agents migration-readiness decision layer for teams still running workflows in Agent Builder — not a replacement for OpenAI's docs, not an API connector, and not affiliated with OpenAI.

WHO

Product, platform, internal-automation and AI engineering teams (and consultants) running visual workflows in OpenAI Agent Builder — including ChatKit embeds backed by Agent Builder-hosted workflows — that still need to move to the Agents SDK (self-hosted) or ChatGPT Workspace Agents before Agent Builder is scheduled to shut down on Nov 30, 2026.

PROBLEM

Deprecation announced Jun 3, 2026; scheduled to shut down Nov 30, 2026; ChatKit remains available. The official export (Code → Agents SDK → TypeScript or Python) then goes into your own app with the Agents SDK, or into a ChatGPT workspace agent (Business, Enterprise, or Edu workspace with workspace-agents access and permission to create agents). OpenAI says it doesn't convert the workflow graph or guarantee every behavior transfers unchanged; strongly deterministic workflows may not migrate faithfully to a workspace agent; connected apps, authentication, publishing, and permissions need separate review in ChatGPT.

SOLUTION

Tick 5 questions — target, workflow count and criticality, connected apps / tools or a ChatKit embed, determinism, export progress → a personal timeline to Nov 30, your path and why, export steps, a manual rebuild list, and a verification & cutover checklist. Never asks for API keys, tokens, org / workflow IDs, workflows, or code; not an OpenAI connector.

RESULT

Before Nov 30, every workflow has its exported Agents SDK code, a chosen home (self-hosted + ChatKit custom server integration, or a Workspace Agent), and a new version verified with representative inputs — instead of finding out in December that the hosted workflow behind your support ChatKit embed is gone.

Features

Features

Marketing points from the product hypothesis — for demand validation, not a formal spec promise.

Timeline risk

days to Nov 30, ChatKit-embed and Evals (Oct 31 read-only) flags

Path picker

Agents SDK vs Workspace Agent, based on OpenAI's guidance

Export steps

Code → Agents SDK → TypeScript / Python, per workflow

Rebuild list

control flow, triggers, tools, permissions, apps & auth, ChatKit backend

Verification & cutover

representative inputs, Preview before Create, staged switch

Path picker (based on OpenAI's guidance)

Your situationSuggested path (per OpenAI's docs)
The workflow is a product feature, needs synchronous responses for end users, or is embedded with ChatKitAgents SDK (self-hosted) + ChatKit custom server integration
Strongly deterministic / branching control flow is the coreAgents SDK (OpenAI: may not migrate faithfully to a workspace agent)
Used by an internal team, maintained in natural language, needs schedules / Slack / sharingChatGPT Workspace Agent (needs Business / Enterprise / Edu + permission to create agents; off by default on Enterprise until an admin enables it)
Triggered from other systems, and you need the result backNote: the Workspace Agent API trigger returns 202 Accepted with no run ID, and the agent's response can't currently be retrieved via the API → lean toward Agents SDK
No ChatGPT Business / Enterprise / EduAgents SDK only

OpenAI's migration guide names only two paths: Agents SDK and ChatGPT Workspace Agents.

Export steps (per workflow)

  1. Open workflow → Code → Agents SDK → TypeScript or Python → copy the complete export. You need access to the workflow.
  2. Save it in your own version control — one workflow at a time, noting which published version it came from.
  3. Agents SDK path: paste the export into your app, install and configure the matching Agents SDK, test it in your runtime, and validate configuration and behavior before deploying.
  4. Workspace path: ChatGPT → Agents → Create → paste the export with OpenAI's prompt Please help me convert this workflow into an agent: → review the behaviors the builder flags for changes.

This page never accepts any pasted export, workflow, or code.

Rebuild list

OpenAI: some workflow behavior may need manual recreation. Review:

  • Control flow and branching, triggers, tools, permissions.
  • Workspace Agent: apps / tools / skills, authentication and connection permissions; end-user account vs agent-owned account (for agent-owned, OpenAI recommends a service account with least privilege); write actions (Always ask by default); Connector Action Constraints; publishing and sharing scope.
  • Agents SDK: runtime configuration, tools, authentication, permissions, deployment; a ChatKit embed moves from the hosted workflow ID to your own server-side agent (ChatKit Python SDK).
  • Evals: Builder Evaluate / trace graders depend on the Evals platform (read-only Oct 31, shut down Nov 30) → set up a separate evaluation flow.
  • Reusable prompts: if you reference them, move the prompt content into your code before Nov 30 (v1/prompts shuts down Nov 30, 2026).

Items named in OpenAI's migration guide and the Workspace Agents help article.

Verification & cutover checklist (static, no credentials)

  1. Take representative inputs from the original workflow.
  2. Run the same inputs through the new version.
  3. Compare behavior: routing / branching, tool calls, output format, refusals and guardrail behavior.
  4. Workspace: Preview first, and Create only after verifying (OpenAI's steps).
  5. Agents SDK: validate configuration and behavior before deploying.
  6. Tools that touch private data or take write actions: keep your existing safety practices.
  7. Cut over in batches; keep the old workflow during the transition window for comparison and rollback.
  8. Finish every cutover before Nov 30.

Never: asking for keys / tokens / IDs / workflows / code, connecting to OpenAI, or running exports or deployments for you.

Join Waitlist

How it works

How it works

Three steps. No OpenAI connection, no keys.

  1. 1

    Answer 5 questions

    target, workflow count and criticality, connected apps or ChatKit embed, determinism, export status.

  2. 2

    Get your path and export steps

    Code → Agents SDK → TypeScript or Python, with OpenAI doc links.

  3. 3

    Rebuild, verify, cut over

    rebuild list, representative-input checks, staged cutover. No keys or code needed.

Use cases

Who it's for

If these situations sound familiar, join the waitlist to help us validate.

ChatKit support embed (hosted workflow ID)

Most urgent — move to ChatKit's custom server integration with an Agents SDK backend, and run both side by side during the transition window.

Internal approval / ticket-routing workflows

A good fit for a Workspace Agent (schedules, Slack, sharing) — but app connection authorization and write-action approvals need to be set up again.

Strongly deterministic multi-step flows (e.g. form validation → branching → internal API call)

OpenAI notes these may not migrate faithfully to a workspace agent → Agents SDK.

Teams with 10+ workflows

Batch by criticality, export one by one, and record versions.

Builder Evaluate as a quality gate

Evals turn read-only Oct 31 — back up how you evaluate and set up a separate flow.

Consultants maintaining workflows for several clients

One path choice and acceptance checklist per client — without ever holding (or needing) a client's keys.

Public signals

Public signals from OpenAI

Official OpenAI pages only — public signals, not endorsements. No customer quotes, logos, or invented metrics.

FAQ

FAQ

Is this an OpenAI tool?

No. Independent; not affiliated with or endorsed by OpenAI.

Is ChatKit shutting down?

No. OpenAI says ChatKit remains available. Agent Builder-hosted workflows in ChatKit are supported during the transition window; OpenAI points migrations to ChatKit's custom server integration.

What happens on Nov 30?

Agent Builder is scheduled to shut down; OpenAI defines shut down as no longer accessible. Finish export and migration before then.

Does export keep everything?

No guarantee. OpenAI says export doesn't convert the workflow graph or guarantee every behavior transfers unchanged.

Agents SDK or Workspace Agent?

Code-first, deterministic, embedded, or needs results back → Agents SDK. Internal, natural-language, scheduled/Slack/shared → Workspace Agent (Business, Enterprise, or Edu with agent-building permission).

Do you need my API key or workflow?

Never. No keys, tokens, IDs, workflows, or code.

Will you export or rewrite for me?

No. The MVP is a static checklist plus waitlist.

When is early access?

We'll email the waitlist in batches; no date promised.

Independent tool. Not affiliated with or endorsed by OpenAI. OpenAI and ChatGPT are trademarks of OpenAI.

Join Waitlist

Waitlist

Join the waitlist

Leave your work email for OaiAgentBld early access and launch updates. Email plus optional checkboxes and dropdowns only — there is no free-text field anywhere on this form.

How workflows are used (optional, multi)
Workflow features (optional, multi)

Email is used only for waitlist, early access and launch updates. Unsubscribe anytime. Never send us API keys, tokens, IDs, workflows, or code.