Get started with brainattic

brainattic is a shared knowledge base your team writes in — and that your AI assistants can read and write too, under rules you set. This guide takes you from an empty workspace to one that answers real questions, with an assistant connected to it.

It takes about twenty minutes.

What brainattic is for

Most teams already write things down. The problem is what happens next: the notes scatter, nobody knows which version is current, and search returns whatever was written most recently rather than whatever is most relevant.

brainattic is for teams at that point — roughly three to twenty people who have outgrown a shared folder. It gives you one place where knowledge lives, retrieval that finds the right passage rather than the newest file, and a way to state your conventions once so that both people and assistants follow them.

If you want an assistant that can genuinely work from your team's knowledge — not one that guesses — this is the part that has usually been missing.

The five things worth understanding

You will meet these in the first ten minutes.

Projects and groups do different jobs, and it is worth getting this right early. A document about a security review of your billing service belongs to the Billing project, and carries the Security group. Someone browsing Billing finds it; so does someone auditing everything security-related across every project.

Create your workspace

1. Go to brainattic.ai and choose Create your workspace.
2. Sign in with Google, or with an email address and password.
3. Give the workspace a name — your team or company name is the usual choice.
4. Accept the Terms and Privacy Policy.

You land on the Dashboard, and your workspace already contains a few seeded documents.

Start with what is already there

Do not start from an empty page. Open Documents in the left sidebar and read the seeded guides — including Welcome to brainattic, which walks through the interface, and the connection guide, which you will need shortly.

These are ordinary documents. Edit them, and they become your own onboarding material.

Write your first document

1. Open Documents and choose the + button.
2. Give it a Title. The Slug fills in automatically and becomes its permanent URL.
3. Pick a Project, or use + Create new project if you have not made one yet.
4. Add any Groups that apply — be generous; they cost nothing and make retrieval work.
5. Write in the Body panel. It is plain markdown, with a live Preview beside it.
6. Save.

Leave Document date empty for anything describing current state. Set it only for historical material — an old email, a retired runbook — where the date belongs to the content rather than to when you filed it.

Every save creates a revision. Open the Revisions tab on any document to see what changed and when, and to restore an earlier version. A save with no changes does not create an empty revision.

Make it worth finding

A few habits pay for themselves quickly:

• Put the answer first, and the background underneath.
• Use a table when every item shares the same fields.
• Link related documents rather than describing them.
• Name an owner and a reviewed date on anything that can go stale.

Search that finds the right passage

Open Search in the sidebar, or press Ctrl+/ from anywhere.

You get three modes:

• Hybrid — the default. Blends exact-term matching with meaning-based ranking.
• Keyword — exact terms only. Use it when you know the phrase.
• Semantic — meaning only. Use it when you know the idea but not the wording.

Results show the project and groups each hit belongs to, a relevance indicator, and the matching passage. You can narrow by project, by group, by time, and optionally include archived documents.

If a hybrid result looks arbitrary on a very small workspace, switch to Keyword. Semantic ranking needs a reasonable amount of content before it settles.

Set the rules your assistants follow

This is the part that makes an assistant useful rather than risky.

Go to Admin → Agent rules. You can write rules at three scopes:

• Tenant rules — apply to every document in the workspace.
• Project rules — apply within one project, and override tenant rules where they conflict.
• Group rules — apply to a group across projects, and override both.

Write down what you would tell a new colleague: where things belong, how documents should be shaped, what vocabulary to use, and what must never be written down. Anyone authoring in that scope sees a banner — "This space has authoring rules" — and assistants can read the rules before they write.

Rules are advisory. They are not enforced by the software, and they are not a security control. Keep them to formatting, vocabulary and authoring guidance — never put credentials, secrets or personal data in a rule document.

Connect an AI assistant

brainattic speaks the Model Context Protocol, so it connects to Claude, ChatGPT, Mistral Vibe, or any MCP-capable client with a single URL.

What you need:

https://app.brainattic.ai/api/mcp

How it works: add it as a custom connector in your client, sign in when redirected, and approve the connection. The client will then be able to search, read and — depending on your permissions — write documents.

Client-by-client instructions are in Connect your AI.

Three things worth knowing before you connect:

• The assistant inherits your permissions. It can do what your account can do, and no more. Connect a read-only account if you only want retrieval.
• Every assistant write is attributed and reversible. It appears in the activity feed under the account that connected, and it creates a revision you can roll back.
• Ask it to read the rules. Rules are advisory, so an assistant will not follow them by accident. Say so in your instruction.

Good first requests: ask it to summarise a document you know, to find something you have not told it the exact words for, or to draft a new page in an area where you have written rules.

Invite a teammate

Open Members in the sidebar and invite by email address, choosing a role.

Roles decide what someone can do — read, write, or administer. Give people the narrowest role that lets them work; it is much easier to widen later than to unwind a bad edit.

How your data is separated

Every workspace is isolated. Content in one is structurally unreachable from another — that separation is enforced on every query, not by convention, and it applies to assistant connections exactly as it does to people.

Within a workspace, what someone sees is decided by their role. A workspace is a trust boundary: everyone in it can see what their role allows across every project in it. If you need two groups of people who must not see each other's material, use two workspaces.

brainattic is built and run on infrastructure Finite Software Systems controls inside the European Union, with customer data processed and stored in Bulgaria. Security has the detail.

Getting help

• Product questions — Support, or email hello@brainattic.ai
• Privacy, data access or deletion requests — privacy@brainattic.ai
• Security reports — privacy@brainattic.ai, the contact published in our security.txt

Never send passwords, tokens or confidential document contents by email. We will not ask for them.

To delete a workspace, or to ask what personal data we hold, write to privacy@brainattic.ai from an address associated with the workspace. Your rights, and how we handle each request, are set out in the Privacy Policy.

Where to go next

Documentation · Connect your AI · Authentication · Security & data · FAQ · Support

Terms · Privacy · DPA · Cookies · Security