What goes into a GPT
A GPT, or custom GPT, is a version of ChatGPT configured for one purpose. According to OpenAI’s help article it can include:
| Part | What it does |
|---|---|
| Instructions | Behavior, tone, goals and boundaries |
| Conversation starters | Example prompts to get people going |
| Knowledge | Uploaded files the GPT uses as reference |
| Capabilities | Tools such as web search or image generation |
| Apps or Actions | Connections to outside services or your own APIs, never both at once |
Your company documents go under Knowledge. A GPT can hold up to 20 knowledge files, each up to 512 MB. Put reference documents in Knowledge and behavioral rules in Instructions, then test whether Preview answers use the intended documents. In a managed workspace you build one like this:
- Open the GPTs area in ChatGPT.
- Select Create.
- Build it conversationally or fill in the editor directly, with instructions and files.
- Test it in preview and save.
Whether you can, depends on product availability, workspace settings and your permissions in Business, Enterprise and Edu. Sharing options, where the admin allows them, include specific people, the whole workspace, a link, or the GPT Store. In workspaces with data residency, GPTs cannot be shared with “Anyone”.
The news: custom GPTs are being retired
As of November 2026, OpenAI says it plans to retire custom GPTs and help creators move their workflows to plugins, which bring reusable instructions, reference files and connected apps together. The dates:
- October 26, 2026: new GPT creation ended for affected Enterprise workspaces (FedRAMP excluded).
- December 11, 2026: standard retirement date.
- February 11, 2027: for FedRAMP workspaces and workspaces with an approved deferral.
Personal accounts (Free, Go, Plus, Pro) can no longer create or publish new GPTs. OpenAI says the transition includes all ChatGPT plans, but creation limits and migration access depend on your account or workspace: follow the notice for yours. It does not say whether Business workspaces share the Enterprise dates. If you are starting now, look at Plugin Creator first.
The common mistake: a GPT learns from your team
A GPT is a fixed setup with fixed files. OpenAI states that GPTs do not use saved memory, custom instructions or previous conversations: every conversation starts fresh. What a colleague corrects in a chat, the GPT will not know next time. Someone has to update the instructions or files. The GPT builder role does not provide access to users’ individual conversations, so ask users to share relevant failures for maintenance.
What it means for your team
A GPT on your documents is a good first step: one place with the right material and the right tone. But it goes stale when the documents change, someone has to maintain it, and now there is a migration to plugins on top. Store your knowledge so it is not tied to one format. For a fixed document set per topic, a shared project is an alternative: see sharing ChatGPT Projects with your team.
Where Bombos fits
Bombos is a plugin-backed memory layer that keeps your team’s knowledge in one place that is not tied to a GPT, a ChatGPT plugin or a tool, in the cloud or on your own server. It reads what already exists (mail, documents, repos), keeps learned rules about how you work, and gives that to ChatGPT, Claude, Copilot and your own agents. Your knowledge survives the next product change. See also how to give AI agents your company context.
Related questions
What happens to my existing GPTs?
They stay usable and editable until their retirement date. After migration to a plugin, the original GPT becomes read-only.
Does a GPT remember past conversations?
No. GPTs do not use saved memory, custom instructions or previous conversations. Each conversation starts fresh.
Can the GPT builder read my chats with it?
No. The GPT builder role does not provide access to users' individual conversations. Workspace administration is a separate role.
What replaces custom GPTs?
Plugins, built with Plugin Creator: reusable instructions, reference files and connected apps in one package. For a fixed set of documents on one topic, a shared project also works.
