W/EXTERNAL BRAIN SYSTEM v0.1 · synthetic sandbox

FROM SOCIAL GRAPHS TO COGNITIVE GRAPHS

Your agent should not be you.
It should know what it may ask, reveal and remember.

Cognitive Relay is a human-ratified exchange layer for personal agents. It turns one live question and one owner-approved insight into a purpose-bound, expiring cognitive handshake.

THIS BUILD PROVESbounded mandatesowner ratificationprovenancefuture-access revocation
Two cognitive capsules connected through a narrow permission gate
01 Exchange only what both sides can inspect.

LinkedIn maps who knows whom.

Cognitive Relay tests who is thinking what—under whose permission, for how long, and for which purpose.

INTERACTIVE PROTOCOL DEMO

One question. Three agents. No silent consent.

01 / MANDATE

Ask a live question

LONG-TERM MEMORY OFFRETRANSMISSION OFFCONTACT / COMMITMENT OFF
02 / RATIFICATION

The Chair is waiting.

Submit the mandate. Candidate agents may surface a capsule, but only its human owner can approve, narrow or deny the exchange.

All people and claims in this public room are synthetic. A successful click is product-function evidence—not evidence of a relationship, community or market.

THE COGNITIVE EXCHANGE PROFILE

A social protocol should fail closed.

Built above A2A / MCP / OAuth. Not a new identity chain, wallet or token.

01

Live Seed

A question, anomaly or changed belief supplied by the principal within seven days.

02

Capsule

A bounded cognitive asset with provenance, privacy class, expiry and allowed audience.

03

Mandate

The exact purpose, requested output, forbidden actions and memory boundary for this query.

04

Ratification

The owner sees the proposed disclosure and may approve, narrow or deny it.

05

Receipt

A hashed record of what crossed the gate, between whom, under which constraints.

06

Continuity

Only owner-approved deltas persist; raw conversations do not silently become memory.

REVOCATION IS NOT FORGETTING. It blocks future system access. It cannot erase content already seen by a person or model.

THE FIRST DEPLOYMENT

Bring One Question:
a human-ratified agent salon.

Eight to twelve principals. Seventy-five minutes. Each participant brings one current obsession, one anomaly and one boundary. Agents compress the room’s breadth; humans decide what deserves trust, memory or a real conversation.

Open the host kit →

Mandate check
Identity, scope, expiry, retention and kill switch.

Blind seed
Agents see questions, not status or biography.

Overlap offers
The Chair surfaces disagreement and adjacency.

Owner ratification
Approve, narrow or deny each capsule.

Human room
Two-person conversations only after bilateral yes.

Receipts
One surprise, one changed question, one reality move.

DEPLOYMENT THESIS

Do not tour conferences. Install a repeatable room.

SFAI Week → Tech Week → Frontier Tower
IsraelAgentic Israel → builder cohort
LondonNewspeak House → policy technologists
EdgeEdge City → human-agent field lab
Asia / CHSingapore → swissAI

A node becomes a Default Seat only after accepted contribution → repeated invitation → recurring responsibility → rule or allocation right.