Does connecting my tools mean the AI can change things in them?

Only if you allow it. In Indexbrain, reads are allowed by default and every write action asks for approval first. Write access is off entirely until an admin turns it on, and there is no auto-approve mode on any plan.

Will my data be used to train a model?

Not by Indexbrain. Your content is never used to train models. Check the terms of whichever model provider you use as well.

Do I have to connect everything?

No, and it is usually better not to at first. One source is enough to tell whether the answers change.

How do I give ChatGPT access to my company's data?

Through a layer that connects to your tools and exposes them to ChatGPT over an API, rather than by pasting the data in or uploading files by hand.

Pasting works for one question and fails on volume, freshness and permissions.

A connector layer reads your tools at the moment the question is asked.

The part worth getting right is who can see what, which a prompt cannot enforce.

The three ways people try this

The first is pasting: copy the thread, the doc and the spreadsheet into the prompt. It genuinely works, and for a one-off question it is the right move. It stops working for three reasons that are structural rather than a matter of effort.

The second is uploading files. Better for a fixed corpus, but it goes stale the moment the underlying document changes, and someone has to remember to re-upload.

The third is connecting the tools themselves, so the AI reads them when the question is asked. That is the only one of the three that stays current on its own.

Why pasting runs out

Volume: you cannot paste thousands of messages and hundreds of pages into a context window, and the answer you want usually depends on more than you can gather.

Freshness: pasted context is stale the moment you assemble it. If the decision changed on Tuesday, the paste from Monday is confidently wrong.

Permissions: a prompt cannot enforce that one person sees a salary discussion and another does not. Whoever pastes decides what the AI sees, which is fine until more than one person is involved.

What connecting actually involves

You authorise a connector to read a tool, usually through OAuth. From then on the AI can search that tool when a question needs it, and cite what it found.

With Indexbrain specifically, ChatGPT connects through a REST API and custom actions, while Claude, Cursor, Claude Code and Codex connect over MCP. The same permissions apply on both paths, including for direct API-key calls.

The free tier includes three integrations and does not expire, which is enough to find out whether an AI that knows your company answers differently from one that does not.

Try it free See it work