Your AI Needs a Second Brain. SGP Now Gives It One.

The problem: every conversation starts from zero

Ask an AI assistant to draft a gas purchase agreement and it will write you a decent generic one. Ask it to draft your agreement, the one with the index price you always use, the take-or-pay terms your counterparties push back on, the clause your lawyer insists on after last year's dispute, and it has nothing to go on. You end up pasting the same background into chat after chat.

That background is most of what makes you good at your job. It's the reason a new hire takes six months to get useful: they have to learn how the company works, who the customers are, which vendors show up on time, what was tried before and why it didn't work. An AI has the same problem, except it starts over every time you open a new chat.

Connecting your AI to SGP over MCP already helps a lot. It can look up your sites, invoices, tasks and contracts directly. But records aren't the same as knowledge. SGP knows that an invoice went out on the 3rd. It doesn't know that this customer always pays late in December and doesn't need a reminder until the 20th. That kind of thing usually lives in someone's head, or in a notes file nobody else can find.

What a second brain is

A "second brain" is an old idea from personal productivity: keep what you learn in notes you can search, link and build on, instead of relying on memory. It works well for people. It works even better for AI, for two reasons.

First, an AI will actually read the notes. People write notes and never look at them again. An AI that's told to check the notes first will check them every single time.

Second, an AI can keep the notes up to date. When you tell it "we switched compressor vendors, Acme was too slow on parts," it can write that down where the next conversation, or the next person's AI, will find it. The notes get better as you work instead of going stale.

So the second brain stops being a chore you maintain and becomes something you and your AI build together while doing the work you were already doing.

SGP now has one built in

Every SGP company now has Second Brain under Back Office > AI Context. It holds plain text notes in Markdown, organized in repos:

Your personal repo. Yours alone. How you like reports written, your own customers and contacts, the things you're tracking. It's created for you the first time you open Second Brain. Company admins can't see it, and if you leave the company you can still download it.
Team repos. One per team, readable only by the members of that team. The field crew's commissioning checklists, the gas marketing desk's notes on counterparties, accounting's month-end routine.
A company-wide repo. What everyone should know: how the company is organized, which sites are active, the standard terms, the people to call.

A new repo comes with a few starter pages, unless you start it empty to import notes you already have: an index.md, a table of contents your AI reads first, and an AGENTS.md that says where things get filed. Your AI reads those, then pulls in only the pages that matter for the question at hand.

Why separate repos matter

It's tempting to throw everything into one big pile and let the AI sort it out. That goes wrong in two ways.

The first is contamination. If the gas marketing team's AI is reading the field crew's maintenance notes, it will start mixing them in where they don't belong. Worse, it might read something it shouldn't: HR notes, a draft deal the rest of the company isn't supposed to know about yet.

The second is bloat. Every page an AI reads uses up its attention. An AI wading through three years of everyone's notes to answer a question about one invoice gives you slower, vaguer answers than one that only sees what's relevant.

Separate repos solve both. Your AI only ever sees the repos you can see: your own, your teams', and the company's. A team repo is closed to everyone outside the team. Admins decide which team and company repos exist, but that doesn't let them read a team repo they aren't a member of, and they never see anyone's personal repo.

You decide what each AI gets

Second Brain tools are part of the same SGP connection your AI already uses, so there's nothing separate to install. When you connect an AI (Claude, ChatGPT, Grok, Gemini or anything else that speaks MCP), you approve it on an SGP screen, and that's where you pick what it can reach. Second Brain is off for a connection until you turn it on, and you can give it read only, or read and write.

That means you can have one AI with full access that you use for your own work, and another, say one wired into a shared tool, that can only read. You can change any connection later, or disconnect it, from AI settings. A connection can never do more than you can do yourself.

It's git underneath

Every repo is a real git repository. Every change, whether you made it in the browser or your AI made it, is a commit with a message saying who made it and through which connection. You can see the history of any page, see exactly what each commit changed, and undo a commit by reverting it. Nothing an AI writes is ever lost, and nothing it gets wrong is permanent.

Because it's git, your notes aren't locked in. You can download any repo as a zip of Markdown files, or as a git bundle you can clone, edit locally with whatever tools you like, and push back. If your AI runs somewhere with a shell, it can do the same.

A few guardrails come along with that:

Secrets are refused. A commit containing something that looks like an API key or a private key is turned away, with the file and line, so it never lands in history.
Notes are data, not orders. Notes often hold text copied from email or chat, and that text can try to tell an AI what to do. SGP tells every AI that reads a repo to treat what's in it as information, never as instructions. Read-only access is the real protection for an AI that doesn't need to write.
Content lives on its own server. File contents and commit messages are stored on a dedicated git server, not in SGP's main database. Search runs there too.

Getting started

Open Second Brain in the side menu and click Get started. Follow the steps for your AI. When you approve the connection, turn Second Brain on.
Ask your AI to "read my second brain and tell me what's there." It will find your personal repo with its starter pages.
Then just work. When you explain something you'd rather not explain again, tell your AI to write it down. If you're an admin, add a repo for each team and one for the company, and seed them with the things new people always ask about.

After a few weeks, you'll notice you've stopped pasting background into chats. Your AI already knows it.