Segmented Brain — memory by subject
Subject breakdown
Agents on this stack
Value delivered
Services
Memories over time (snapshots every 60s)
Memories per project
How to read this
Counts come from logged agent turns (
conversation_turns) and the
compression cache. When the Headroom proxy is up, its live counters are the
source of truth — run headroom savings for the audited number.
Curator — expiry sweep · duplicate report · project briefs
Recent runs
Autopilot Engine — self-heal · cleaning · maintenance
Value Dashboard — lifetime impact
Recent activity log
Alerts
Waiting for a person — forget requests and account-wide memories
An agent forgets its own memories at once. Forgetting what another agent or a
person wrote, and putting an agent's memory in front of every project, waits here. Every decision
is on the audit ledger. Also:
bin/brain approvals.Agent calls — last 24 h
Request inspector — what each agent sent and got back
Secrets are redacted before anything is stored. Click a row for the full call and
the changes it made.
Connect a drive
Anything the stack can read from. Linking a drive never changes it — ingest runs are read-only and leftover files are only reported.
Train the brain
Consumed vs leftover
Why files were left behind
Subjects learned
Run history
Modules connected over the API
What each agent can reach, and whether it is actually being used. A module with zero calls is dead weight; a module that is failing is worth knowing before the agent silently degrades.
Subagent lineage
Depth matters more than count: twenty subagents at depth one is a very different shape from five nested four deep.
Subagent detail
Pick a subagent to see the memories it produced, or move it under a different parent. Relinking moves the pointer — no memory is ever re-ingested or lost.