Team Brain: one source of truth, one manager, every agent and teammate
Team Brain gives every agent and teammate one shared, up-to-date source of truth about your team, kept by one agent responsible for it.

When our team's brain first read our docs and apps, it found six content calendars in Notion. Only one was still in use. It also found our copy rules split across several agents and a Notion doc, and they contradicted each other. Each agent writing our copy followed its own rules, and nobody had decided which ones were current. I'm a founding engineer at Gamut, and the team and I have spent the last few weeks building Team Brain.
None of that was unusual. It's what happens to any team's context once it lives in many places. Docs drift apart. Teammates keep different versions in their heads. Each agent learns its own copy, and nobody knows which one is current.
Agents make this worse, not better. An agent acts on whatever context it has, with full confidence. If its copy is out of date, it gets the answer wrong and doesn't know it.
A team needs one source of truth. That's hard to keep across many teammates and many agents, so it needs a simple building block.
Team Brain is that building block:
One source of truth, one manager responsible for it, shared by every agent and teammate.
What Team Brain is
Team Brain is two things: one Team Brain agent and one Team Brain drive.
The drive holds the brain itself, a folder of plain pages about your team. It covers who the team is, what it sells, how work gets done, what it has decided, and where everything else lives.
The agent is the manager. It builds the brain, keeps it up to date, and answers questions about it.
Both are shared with your team automatically. Every teammate can chat with the Team Brain agent. By default, every agent in your workspace reads the brain.
This is different from an agent's own memory, which stays with that one agent. It's also different from search, which digs through your raw docs every time someone asks, conflicts and all. The brain is built up and kept current over time, and everyone reads the same one.

What it's built on: drives
For anyone new to Gamut, Team Brain is built on a feature we call drives.
A drive is a store of files that lives outside any one agent or teammate. It can hold context, files, and data. You add a drive to an agent to give that agent access.
A drive can live in Google Drive, Dropbox, GitHub, an S3-compatible bucket, or a Gamut drive that we store for you. In the desktop app, it can also be a folder on your computer.
OK, how does a drive work?
We do a lot of work behind the scenes so that a drive is mounted as a folder on the agent's computer. The agent sees the drive as a normal folder. It reads and edits files there the way it would any other files.
That keeps the agent's job simple. It doesn't need to know whether a file lives in Google Drive or GitHub. Agents, like people, work better when their load stays low. We try to give them tools that match what they already learned in training, and files are one of those.
Here's what happens when one agent learns something:
- Agent A writes a new fact to a page in the drive.
- The drive saves it to wherever it lives, like Google Drive.
- Agent B, which has the same drive, reads the page.
- Agent B now knows the fact too.
Who can read and write
When you add a drive to an agent, the agent can read and write by default. You can make it read-only for any one agent. That fits context you want an agent to follow but never change, like steps only people should edit.
A drive can be private, or an admin can share it with everyone on the team. Teammates can then add a shared drive to their own agents. An admin can also make a shared drive global, which gives it to every agent, including agents made later.
You can use drives however fits your team: company context, best practices, how you work. Team Brain is the way we found drives most useful for teams.
Why one manager
A shared drive that every agent writes to has a problem. Over time it fills with duplicate pages and facts that contradict each other.
So the Team Brain drive works differently. Other agents read it but don't write to it by default. The Team Brain agent is the manager, and it makes the changes. When another agent learns something, it sends it to the Team Brain agent, which decides where it belongs.
One manager means one page per fact, one structure, and one record of every change.
Filling it
An admin sets up Team Brain from the Team Brain tab in the sidebar. Setup has four steps:
- Choose where it lives. Pick the drive that holds the brain.
- Choose its agents. Pick which agents read it, or turn on "Every agent".
- Set a schedule. Pick how often upkeep runs: hourly, daily, or weekly.
- Context ingestion. Start a chat with the Team Brain agent to build the brain.
In that chat, the agent researches your team's tools first. It reads the tools you connect, like Slack, Notion, Google Drive, GitHub, Linear, Jira, or Gmail. Then it asks you only what those tools couldn't answer.
The first build is a good time to find where your docs disagree. We tested it on a brain built from a made-up company's docs, with three conflicts planted in them. The first build flagged all three, and found a fourth we hadn't planted. The more sources you connect, the fuller the picture.
Keeping it up to date
Upkeep runs on the schedule you set. Each run, the agent rereads your sources, updates the brain, and writes a short summary on the Team Brain page.
Each summary has three parts:
- Highlights: what changed for the team.
- Changes: which pages it added or updated.
- Needs you: questions it can't answer alone, split into "Blocked" and "To decide".
The agent follows a few rules that keep the brain honest:
- Every fact has a source. No evidence, no change.
- Every fact has a label: verified, reported, or inferred.
- Nothing gets deleted unless a person asks. Old pages go to an archive.
- A person's edit wins. If a teammate changes a page by hand, the agent keeps it.
- Every change is logged, with a note on how to undo it.
When sources disagree, it doesn't pick one. Two of our records once disagreed about whether some partner terms were in writing. The agent marked the page "records disagree" and asked the teammate who owns partnerships.

These are defaults. Because the manager is an agent, you change how it works the way you'd brief a teammate: you tell it. Ask it to reorganize a folder, stop reading a source, or check with you before filing anything about customers. It updates its own instructions and works that way from then on.
Growing it with your team
Each teammate holds context the rest of the team doesn't have. An engineer knows how the product works. Someone in sales knows what customers ask for. Most of that has never been written down in one place.
Team Brain gives each person one place to add it. There's one Team Brain per workspace, which for most teams is the whole org. Once an admin sets it up, every teammate can chat with the Team Brain agent and add what they know. They can also add sources only they work in, like the Notion pages where growth plans campaigns, or the doc where engineering keeps its coding principles. Teammates turn on Team Brain in Settings, then Experiments, to see it.
A correction from one person reaches every agent. In October, our brain recorded a new sales target as secondary. Our CRO said that was wrong: it was the primary target. The page was fixed that day. Then one of our sales agents flagged one leftover line that still said "secondary", and that was fixed too.

This helps most when someone from a different side of the team joins. Their tools and knowledge feed the brain, and everyone else gets that context.
Because the Team Brain agent is a normal Gamut agent, it can do what any agent can. Other agents can message it. You can ask it anything about the team.
How it has helped us
Our own brain has run for about a month now. It has 99 pages and a changelog of 767 changes. It reads our GitHub, Linear, Notion, Google Drive, and both of our Slack workspaces.
Every night it posts a summary. I read it, and when it asks me something under "Needs you", I answer in chat and the brain updates. Teammates do the same. Our product lead asked the Team Brain agent for one home for our logos, and it made the logos page and folder. Each summary also lists what it left out on purpose, like customer details and figures.
While planning this post, I asked my own agent where Gamut publishes, who runs our content calendar, and which copy rules apply. It answered from the brain in a few minutes, with links, without asking anyone on the team.
What it doesn't do yet
Team Brain is still experimental, and we're still improving it.
- It only knows what you connect it to and tell it.
- Its labels are its own judgment, so check anything important against the source.
- The brain lives in a drive. An Obsidian vault works today as a folder on your computer, in the desktop app. Notion can be a source, but not the brain's home yet.
- Each workspace gets one Team Brain.
Turn it on
Go to Settings, then Experiments, and turn on Team Brain. If you're an admin, open the Team Brain tab and click "Set up Team Brain".