Sharing memory between agents without sharing everything

Teams of agents work better when they share context. A research agent finds the facts, a writing agent uses them, a review agent checks both. The risk is the opposite extreme: one big shared memory where every agent sees everything, including notes it should never act on.

One mind per scope

In Pinlo a mind is the unit of sharing. Keep one mind per project, client or strategy, and share minds rather than individual records. An agent that works on project A never receives a grant to project B's mind, so a recall can never mix the two.

One key per agent

Each agent signs with its own key. That makes every record traceable to the agent that wrote it, and it means revoking one agent does not disturb the others. Never give an agent the owner's wallet: the owner key is the one that can grant and revoke.

Read, or read and write

Most agents only need to read. Give write access to the few that should add to the shared record, and let the rest consume it. On-chain, PinloCortex checks the grant before accepting a new version from a writer, and it refuses a version built on an outdated one, so two writers cannot overwrite each other silently.

Expiries do the cleanup

A grant can carry an end time. Contractors, short experiments and temporary review agents get access that ends on its own, with nothing to remember to revoke.

Grants are public, text is not

On-chain grants show which addresses can read which mind. The memory text itself stays off-chain, and anything sensitive can be encrypted in the browser before it is saved. Plan labels with that in mind: a label is visible to anyone reading the chain.

Try a grant in the app: open a mind, add a record, and share it with a second address you control.