Skip to main content

​
One Store per Repository

Brain stores repository memories in one local store. Searches include memories recorded on different branches by default. Each memory keeps references to the branch, session, and sources it came from. An explicit branch filter selects memories with matching branch references. It does not create a separate store. Migration combines records with the same memory ID and keeps their source references and status. Memories with different IDs remain separate, even when their text is similar. For backup, migration, and automatic cleanup, see Upgrade and Recover.

​
Global Memories

Global memories use a separate local store and are available from any repository. Repository hosted sync does not upload this store. See Save a Rule for Every Repository for commands to save, list, and search global memories.

​
Connect a Repository

Connect the local repository to its hosted Entire repository:
Connecting requires Entire CLI 0.11.6 or later. It records the target and attempts a sync immediately. After connection, Brain can sync when memory extraction finishes, when you commit or push, and during background maintenance. Memory sync sends repository memory records across branches and receives the team’s shared memories. It does not upload the complete local Brain directory or the separate global store. Session and checkpoint storage have their own configuration. To request an immediate sync:
To disable automatic hosted sync for this repository:
A plain memories sync fails when the repository is not connected. Memories remain local unless you connect or supply an explicit manual target.

​
Use an Explicit Target

Scripts can provide --repo-id, --api-url, and --token to entire brain memories sync. Use --memories-backend=http for an unconnected repository. For a manual sync, explicit flags take precedence over the matching environment values and the recorded connection. Background sync uses the recorded connection; these overrides do not redirect it. See the security policy for the full target and credential rules.

​
Network Controls

Set ENTIRE_BRAIN_NO_EGRESS=1 to block Brain-controlled network operations:
This blocks hosted memory sync, webhooks, non-local model calls, and repository cloning through Brain. It takes precedence over connection state and target flags. Selecting --agent none or a local model alone does not disable hosted memory sync. A custom --agent command runner executes a local program. No-egress mode does not restrict that program’s own network calls. Memory extraction and automatic category selection can send context to a model provider and incur token charges. Webhooks stay off until you configure them. See the webhook guide.

​
Concurrent Writes and Background Work

Your commands take priority over queued background writes. Brain finishes the current write before starting the next one. A large migration can still make another command wait. entire brain stop pauses background work. Other agents can still run commands that write memories. Finish active writes and stop other writers before making a backup. Resume paused background work with entire brain start.

​
Mixed Client Versions

Brain v0.5.1 can exchange hosted memories with stable v0.2.0, v0.3.0, v0.4.0, and v0.5.0 clients. Each client must use a separate local store. Brain v0.2.0 to v0.4.0 cannot read a migrated local store. Do not share the upgraded directory with those clients.