BigBrain is a tool for building life-wide context for agents.
You drop stuff in (webpages, papers, notes), or connect sources (like email or Granola).
Then, your agents magically have access to a bunch of context about your life!
Items enter your Vault.
Then, you connect clients like Claude Code or Codex.
All of your agents will have context about your life. (They can also add items to your Vault).
Integrations let you auto-pull from outside sources. You can use existing integrations or build your own.
From inside BigBrain, you can use the Pilot, BigBrain's in-app agent.
If you're a power user, you can also use the Pilot to orchestrate agents.
If you're like me, you run a lot of agents at the same time. Running them through BigBrain can be very powerful.
That’s how BigBrain works at a high level.
But the magic of BigBrain comes from the Vault. Let’s zoom in on it to see how it works.
Items enter the Vault as-is. BigBrain never edits sources—it always keeps the original.
The BigBrain Vault turns those sources into a network, or Graph.
The Graph is made of sources—the items you added…
…and entities—things the model extracts from those sources. Entities might be people, places, projects… anything you might say a source is “about,” or “concerns.”
The Graph is built by the Gardener, a model that reads each source and creates assertions about it:
In the [9-23-26 Brain Exercise meeting], [Nick Merrill] committed to creating a visual explainer for [BigBrain].
The Graph is a projection of these assertions.
You may change the Gardener model over time. No worries. Assertions show which model authored them.
At night, the Memory model reviews recent assertions and uses them to update Memories.
Memories help agents (and you!) find their way around. You can think of Memories like signs marking the “regions” of your Vault.
These models run on your existing AI subscriptions.
I never see your Vault. Vault data goes only to your model provider.