---
title: Brains
description: Brains are isolated memory workspaces. One per project, one per environment, or one per agent — your call.
order: 5
---

# Brains

A *brain* is an isolated memory workspace inside your Memoria account. All memories — episodes, entities, edges, playbooks — live inside exactly one brain. Brains never share data; recall in brain A never returns facts from brain B.

## Why brains exist

You'll usually want more than one brain. A few common shapes:

- **One brain per project** — your coding agent's `acme-app` brain stays separate from your `personal-notes` brain.
- **One brain per environment** — `production` brain vs `staging` brain so test data doesn't pollute live recall.
- **One brain per persona** — a customer-support agent's memory of one customer kept isolated from another's.

Inside one brain, everything is connected: entities are resolved across episodes, the knowledge graph is dense, recall is fast. Across brains, nothing leaks.

## Creating a brain

You get one brain (called `default`) automatically when you sign up. To create more, go to **Dashboard → Brains → New brain** and pick a slug. Slugs are unique per account and used in URLs and API calls.

## Binding keys to brains

Every API key is bound to exactly one brain at creation time. The key inherits the brain — you can't switch later. To rotate brains, generate a new key. This makes leaks safer: a leaked key only exposes one brain's data, never the whole account.

For MCP/OAuth, the brain is chosen at consent time (the picker on the authorization screen). The access token carries that binding for its 1-hour lifetime; refresh tokens stay bound to the same brain.

## Deleting a brain

Deleting a brain is irreversible. It cascade-deletes:

- All episodes, entities, edges, and playbooks in the brain
- All API keys bound to the brain
- All OAuth refresh tokens scoped to the brain (active access tokens expire naturally)
- All search indexes and stored data for the brain

There's a 30-day account-deletion grace period for the brain itself (so accidental deletes can be recovered by support), but the *contents* are wiped immediately. If you need a longer grace period, contact us before deleting.

You cannot delete your last remaining brain — Memoria always keeps at least one brain per account so the dashboard has somewhere to render.

## Brains vs tenants vs accounts

A bit of terminology, useful to know:

- A **tenant** is the billing unit (your Memoria account).
- A **brain** is an isolation unit inside the tenant.
- Every stored record is keyed by both tenant and brain — reads are scoped to both, always.

For most users, tenant and account are interchangeable. Multi-seat workspaces (one tenant, multiple human users) are launching with the Team plan.

## Best practices

- **Start with one brain.** Don't over-partition. You can always add more when the need is concrete.
- **Name them descriptively.** `acme-prod` beats `b1`. The slug shows up in dashboard URLs.
- **Don't put unrelated projects in one brain.** Entity resolution will try to merge them (e.g., two unrelated "John"s) and recall results get muddled.
- **Use environment brains in CI.** Spin up a brain per CI run if your tests need a clean slate — delete it at the end.