How it works

1. Make a brain for an area of life. Work and family are the two you start with. Add another when the rules or the secrets are different. 2. Connect agents to that brain only. Each agent has a name, a provider, and one delivery mode: push, pull, or relay. 3. Send a packet. The brief is self-contained. You can require evidence, and you can require your approval before the packet counts as done. 4. The packet lands in that agent's inbox. Push wakes a webhook. Pull waits for check-in. Relay gives you a signed link. 5. The agent posts a result. Without evidence, iBrains refuses it. If the packet was gated, it sits in that brain's approval queue until you approve or reject. 6. The audit trail records the create, the delivery, the result, and the decision.

A worked example

You write "Review the release notes" and address it to Claude on the work brain. Claude is pull, so the packet waits in Claude's inbox until the next check-in. From that chat you say "send the follow-up to Buddy." Buddy is push, so iBrains signs a webhook and Buddy's bot picks the packet up. Buddy's result includes the evidence you asked for. The packet said publish needs approval, so it stops in the work approval queue. The family brain never sees the packet, even if an agent there asks.