In short
- ChatGPT with files is great for one person, one project, a few documents that don't change.
- It breaks when a team needs the answers: copies go stale, nobody knows the source, everyone sees everything.
- A company brain reads your tools, keeps a living file per client with a source for each fact and answers each person within their permissions.
- You don't have to choose an AI. Your team keeps ChatGPT or Claude and connects it to the brain.
Every team that takes AI seriously goes through the same phase. Someone creates a ChatGPT project, uploads the price list, two proposals and the onboarding doc, and for a few weeks it feels like magic. Then a client asks about a discount that was agreed by email last Tuesday, and the project has no idea.
That isn't a ChatGPT problem. ChatGPT is very good at reasoning over what you give it. The problem is what you give it: a pile of copies, frozen on the day someone uploaded them, with no idea who is asking.
Where ChatGPT with files works well.
Be honest about this part, because for a lot of work it's all you need. Uploading files to a chat or a project is the right tool when:
- One person is working on one piece of work, like a proposal, a report or a contract review.
- The documents are few and they don't change during the work.
- Nobody else needs the answers, so permissions don't matter.
- Nothing has to happen afterwards. You read the answer and act yourself.
If that describes how your company uses AI, keep doing it. You don't need anything else yet.
Where it breaks for a team.
The trouble starts when the same files have to serve five people and three months of changes. Here is the same question, asked the same morning, to a ChatGPT project and to a company brain:
What price did we agree with Northwind?
According to proposal_v2.pdf, Northwind pays $1,100 a month.
What price did we agree with Northwind?
$1,200 a month from June, with 10% off the first three months. Leo agreed it by email on September 2.
Neither answer is "AI being wrong". The left one is correct for March. It just doesn't know March is over. In a team, that pattern repeats in four ways:
- Copies go stale. A file uploaded to a chat is a photo of the day it was uploaded. Prices, contacts and agreements keep moving in Gmail, HubSpot and Slack.
- No source per fact. You get an answer, not the email it came from. Before you send it to a client, you still have to go and check.
- Everyone sees everything. A shared project has no idea that Sales shouldn't see margins or that only the founders see salaries. So either you upload less, or you share more than you'd like.
- Everyone re-explains the company. Each person builds their own little project with their own copies. Five people, five versions of the truth.
What a company brain actually is.
A company brain flips the direction. Instead of people uploading copies into a chat, it reads the tools where the company already writes things down, and keeps what it learns organised and current.
In Brainstack that means three things:
- It reads only what you connect. Gmail, Google Drive, HubSpot, Notion, Slack, Microsoft Teams and the rest of the tools you switch on. Nothing else.
- It keeps a living file for every client. Status, decisions, commitments and open questions, each fact with a link to where it came from.
- It stays up to date on its own. Every weekday at 19:00 it brings active clients up to date, and once a week it looks for information that has gone stale.
Where it lives.On a server only your company uses, with encrypted backups every 6 hours. Your AI provider processes what a person or an agent sends it, under the terms of your plan.
The part nobody plans for: permissions.
The moment more than one person asks the same AI about the company, "who can see what" stops being an IT question. A company brain answers each person with only what their role allows. Try it on the same client file:
As the founder, you see all four facts.
From answers to work.
There's a second difference that matters more over time. Files in a chat can only answer questions. A company brain is also where your agents work from.
The billing agent reads Northwind's payment terms before chasing an invoice. The follow-up rep reads what was promised in the last call before drafting the email. The support agent reads the docs that apply to that client. Each one works on a schedule and asks before anything leaves the company. If you're thinking about your first one, start with how to hire your first AI agent.
And your team doesn't have to change tools. With team AI access, in beta today, each person connects the Claude, ChatGPT or Cursor they already use to the brain. It covers knowledge, tasks and projects, always within that person's permissions.
Connect two tools and the brain builds the first client files the same evening.
Side by side.
| ChatGPT with files | Company brain | |
|---|---|---|
| Where knowledge comes from | Files someone uploaded | The tools you connect |
| Stays current | Until someone re-uploads | Every weekday at 19:00 |
| Source for each fact | The file name, at best | Link to the email, deal or doc |
| Per-person permissions | Whoever has the project | Per person, role and space |
| Agents can work from it | No | Yes, with approvals |
| Your team's own AI | One app | Claude, ChatGPT or Cursor (beta) |
| Best for | One person, one project | A team, every client, every week |
Which one do you need?
- It's mostly you using it
- Your documents rarely change
- Nobody needs different permissions
- You act on the answers yourself
- Three or more people ask about the same clients
- What was agreed lives in email, CRM and chat
- Some things shouldn't be seen by everyone
- You want agents to do the work, not just answer
Most small companies cross that line earlier than they think, usually around the third person and the tenth client. The good news is you don't have to throw anything away: the brain reads the same Drive folder you've been uploading from, and your team keeps the AI they like.