Will my custom GPT work with Claude?

Its instructions will, if you rebuild it as a skill in a tool that runs any model. Check the instructions read well for both families: spell out every step for literal GPT models, and state clear limits for Claude, which tends to fill gaps.

How do I keep a custom GPT's knowledge files up to date?

In Interweave OS, a skill's documents live in a folder in Files. Drop a newer file into that folder and the skill uses it from then on, with nothing else to edit.

Who can use a custom GPT once it is moved?

In Interweave OS you choose when you save the skill: everyone in the organisation, one team, or just you. Nobody has to pick it from a list; the agent uses it whenever a request matches its description.

How to move your custom GPTs out of ChatGPT

Q: Can I export a custom GPT from ChatGPT?

You can copy everything that makes it work. Open the GPT, choose Edit, then Configure, and copy its name, description, instructions and conversation starters, and keep the files you gave it as knowledge. Those are all you need to rebuild it somewhere else.

Custom GPTs are one of the best things people have built with AI at work. A brand voice checker, a tender reviewer, a policy helper that knows your handbook. But they only run inside ChatGPT, on OpenAI’s models. If your organisation wants to use other models too, or wants the whole team to share that know-how in one place, it is time to move them. The good news: it takes minutes, and you lose nothing that matters.

Custom GPTs retire on 11 December 2026. OpenAI is replacing them with plugins, and custom actions do not carry over. What is happening and what to save now.

What a custom GPT really is

Strip away the icon and the name, and a custom GPT is four things:

A name and a description: what it is and when to use it.
Instructions: the standing guidance it follows every time.
Conversation starters: example requests that show what it is for.
Knowledge files: the documents it reads from, such as a style guide or a price list.

All four can be copied. Any tool that lets you save standing instructions with documents attached can run the same job. In many tools, including Interweave OS, that is called a skill: a set of instructions, plus documents if you like, for one kind of job.

Before you start: take stock

If you have built a lot of GPTs, do not move them all at once. Make a quick list first.

Which GPTs do people actually use each week?
Which ones are personal, and which ones does the whole team rely on?
Which ones have knowledge files that go out of date, such as price lists or policies?

Start with the two or three that matter most. The rest can follow, or retire quietly.

Step 1: open the GPT’s settings in ChatGPT

In ChatGPT, open your GPT, choose Edit, then Configure. This shows everything that makes the GPT work. Keep this tab open while you copy.

Step 2: make a new skill

In the tool you are moving to, start a new skill (or whatever that tool calls saved instructions). In Interweave OS, open Skills and make a new skill.

Step 3: copy the name, description and instructions

Copy the GPT’s Name, Description and Instructions into the same boxes.

Take a minute over the description. In ChatGPT, people pick a GPT from a list, so a vague description did no harm. In a tool that picks the skill for you, the description is what tells the agent when to use it. A vague one means the skill gets missed, or used for the wrong job.

Say what it is for and when to use it:

Weak: “Tender stuff.”
Strong: “Use when reviewing a government tender response before submission: checks every mandatory criterion is answered and flags missing attachments.”

Step 4: add the conversation starters as examples

Conversation starters are short, real examples of what the GPT is for. Add them to the skill as examples. They help the agent recognise matching requests, and they help your colleagues see at a glance what the skill can do.

Step 5: upload the knowledge files

Upload the files you gave the GPT as its knowledge. If you no longer have the originals, find them in your shared drive before you start, so you upload the current version rather than an old copy.

Step 6: choose who can use it, and save

Decide who should have it:

Everyone in the organisation, for house style, policies and anything company-wide.
One team, for a team’s own process, such as the sales quote checker.
Just you, for a personal helper that only suits the way you work.

Then save.

Make the instructions work for any model

A custom GPT was written for OpenAI’s models. Once it is a skill on a model-neutral platform, it may be run by a Claude model one day and a GPT model the next. The two families read instructions differently, so a quick tidy-up pays off.

GPT models are literal. They do exactly what is written. Number the steps, name every file, and say what done looks like.
Claude models fill gaps. They work out what you probably meant and often do more than you asked. State the limits: what it must not change, what is out of scope.

If you are not sure, ask the agent to do the tidy-up for you.

Try saying:

Read the instructions in our Brand Voice skill. Rewrite them so either a GPT or a Claude model would follow them well: number the steps, say what a good result looks like, and list what it must never change.

Test it with the old conversation starters

Before you tell the team, try the skill with the same requests people used to send the GPT. You do not need to pick the skill. Just ask, and check that the right skill is used and the answer is as good as before.

Try saying:

Rewrite this paragraph in our brand voice.

If the result misses, look at the description first. Most problems come from a description that does not say clearly when to use the skill.

It is also worth running the same request on both model families and comparing. You may find one suits this job better, or that a small change to the instructions gets a better result from both.

Knowledge that keeps itself current

The weakest part of most custom GPTs is the knowledge. Someone uploads the price list in March, and by June the GPT is quoting old prices. Nobody remembers to update it.

A better setup keeps a skill’s documents in a normal folder that people already look after. In Interweave OS, a skill’s documents live in a folder in Files. Drop a new PDF, price list or policy into that folder and the skill uses it from then on, with nothing else to edit. You can also make the folder read-only for the skill, so it can read your material but never change it.

Bring the rest across too

Custom GPTs are not the only thing your other AI tools know about you. Over months, they pick up your role, your preferences and your projects. In Interweave OS, the Memory page has Import from your other AIs: one prompt to paste into ChatGPT, Claude, Gemini or any other AI you use. Paste their answers back, and Interweave OS sorts them into your memory, your projects and your skills, custom GPTs and Gems. You tick what to keep before anything is saved.

Worked something out in a single ChatGPT conversation? Make a share link there, paste it into a new chat, and ask the agent to read it and carry on.

Try saying:

Read this ChatGPT conversation: (paste the share link). Pick up where it left off, and list what we decided.

A short checklist

List your GPTs and pick the two or three that matter most.
For each one: copy the name, description, instructions and conversation starters, and upload the knowledge files.
Rewrite the description so it says when to use the skill.
Tidy the instructions so either model family can follow them.
Choose everyone, a team or just you.
Test with the old conversation starters, on both families.
Put the knowledge files in a folder people already keep up to date.

How Interweave OS does it

In Interweave OS, a custom GPT becomes a skill: open Skills, make a new skill, copy across the name, description, instructions and conversation starters, upload the knowledge files, and choose everyone, a team or just you. The agent then uses it whenever a request matches, on whichever model the chat runs, Claude, GPT, Gemini or another. Everyone skills work in every chat, and team skills in that team’s chats. Private skills work in chats where you are the only person; once a chat is shared, it uses everyone and team skills only. The skill’s documents live in a folder in Files, so dropping in a newer file keeps it current.