Can I use a non-OpenAI model in a Custom GPT?

No. Custom GPTs run on OpenAI's models. In Relaite you choose the model per agent from anything available on OpenRouter, so a cheap model can handle chatter while a stronger one reads your contracts.

Do the people I share with need an account?

A Custom GPT lives inside ChatGPT, so anyone you share it with needs a ChatGPT account. A Relaite agent is shared with contacts inside Relaite, and they use it in their conversation with you.

Can people see the instructions I wrote?

In Relaite they cannot: someone you share an agent with sees its replies, never its instructions. Custom GPT instructions are stored with OpenAI, and there is a long public history of users extracting them from shared GPTs with ordinary prompting.

Which is better for questions about a spreadsheet?

They work differently. Code Interpreter writes and runs Python, which is more flexible and can do things Relaite cannot. Relaite stores a CSV as a table and runs the model's SELECT against it in sandboxed read-only SQLite on every relevant question, so the arithmetic is done by the database rather than the model, every time, with no sandbox to spin up.

Relaite vs Custom GPTs

By Örvar Karlsson, founder of Relaite · Published 29 July 2026 · Last updated 29 July 2026

These are closer rivals than ChatGPT itself — both let you configure an assistant once and reuse it. Custom GPTs have the deeper toolbox: Code Interpreter runs real Python, Actions call any API you like, and the GPT Store puts your work in front of strangers. Relaite trades that reach for three things: any model per agent, your own key, and instructions that stay yours when you share.

At a glance

	Relaite	Custom GPTs
Where it lives	Its own app (web, installs on phone and desktop)	Inside ChatGPT
Models	Any model on OpenRouter, chosen per agent	OpenAI's models
How you pay	Free app; you pay OpenRouter directly at their rates	Through your ChatGPT plan
Spreadsheets	CSV stored as a table, queried with SQL on every relevant question	Code Interpreter writes and runs Python
Documents	Per-agent knowledge base with citations	Knowledge file uploads
Calling your own APIs	No — a curated skill catalogue only	Yes, via Actions
Sharing	Named contacts; they @mention it in your chat	Link or the GPT Store
Do recipients need an account?	A Relaite account	A ChatGPT account
Are your instructions private?	Yes — recipients see replies only	Stored with OpenAI; extraction is a known issue
Public distribution	None	GPT Store

Three real differences

1. One vendor's models, or all of them

A Custom GPT runs on OpenAI's models. That's a good default and a hard ceiling. In Relaite the model is a per-agent setting over anything OpenRouter offers, so a throwaway agent can run something cheap while the one that reads your contracts runs something strong — and when a better model ships next month, you switch that agent and nothing else changes.

It also means the bill is yours. You use your own OpenRouter key, pay their rates with no markup from us, and see the spend in your own dashboard.

2. Your instructions stay yours

The prompt you tuned over an afternoon is the actual work. Share a Relaite agent with a contact and they get its replies — never its instructions. Custom GPT instructions live with OpenAI, and there is a long, well-documented public history of people extracting them from shared GPTs with nothing more than persistent prompting.

If what you built is worth sharing, it's worth not giving away.

3. Spreadsheets: deterministic, not clever

This is the difference I'd push back on hardest if I were you, because Code Interpreter is genuinely more capable. It writes Python, so it can reshape, join, plot and model in ways Relaite simply cannot.

What Relaite does is narrower and always on. A CSV you attach is stored as a table; when a question needs it, the model writes one SELECT and Relaite runs it in a throwaway read-only SQLite database — single statement, row-capped, five-second deadline. The arithmetic is done by the database, on every relevant question, with no sandbox to spin up and nothing to re-upload next session. For "what did we close in December" and "sum this by month", that's the whole job.

When Custom GPTs are the better choice

Pick Custom GPTs if…

You want Code Interpreter's full Python sandbox
You need Actions to call your own APIs
You want strangers to find your assistant in the GPT Store
You already pay for ChatGPT and so does everyone you'd share with
You want voice, image generation and native mobile apps

Pick Relaite if…

You want a different model per agent, on your own key
Your instructions shouldn't travel with what you share
Your questions are mostly exact lookups over your own files
The people you'd share with aren't ChatGPT subscribers
You'd rather pay per message than hold a subscription

The honest summary

Custom GPTs are the better product if you live inside ChatGPT already and want maximum capability per assistant. Relaite is the better fit if the assistant needs to run on a model you chose, on a key you own, and be handed to specific people without handing over how it works. Neither of those lists is a knockout — pick the one whose constraints you can live with. And if you want the general-assistant comparison instead, that's Relaite vs ChatGPT.

Common questions

Can I use a non-OpenAI model in a Custom GPT?: No. Custom GPTs run on OpenAI's models. In Relaite you choose the model per agent from anything available on OpenRouter, so a cheap model can handle chatter while a stronger one reads your contracts.
Do the people I share with need an account?: A Custom GPT lives inside ChatGPT, so anyone you share it with needs a ChatGPT account. A Relaite agent is shared with contacts inside Relaite, and they use it in their conversation with you.
Can people see the instructions I wrote?: In Relaite they cannot: someone you share an agent with sees its replies, never its instructions. Custom GPT instructions are stored with OpenAI, and there is a long public history of users extracting them from shared GPTs with ordinary prompting.
Which is better for questions about a spreadsheet?: They work differently. Code Interpreter writes and runs Python, which is more flexible and can do things Relaite cannot. Relaite stores a CSV as a table and runs the model's SELECT against it in sandboxed read-only SQLite on every relevant question, so the arithmetic is done by the database rather than the model, every time, with no sandbox to spin up.

Try Relaite free