7. Knowledge & Organizational Memory (GBrain)

Story moment

Memory stops being a person

GBrain matters because repeated context should become system memory, not another tax paid by the operator.

Chapter 7

Executive Summary

What this chapter establishes

GBrain and Knowledge let the organization stop asking humans to repeat context.

Operator ExperienceAgents arrive with prior decisions and cite evidence instead of asking for the same context.

Concept trail

Role Replacement Work Order Reference Realization Receipt Authority

Autonomous organizations require memory. Without it, the system repeatedly asks the human to re-explain the same thing.

**Knowledge and Memory Flow***What to notice:* Notice memory sits between sources and reasoning, so agents arrive with precedent instead of asking again.

Knowledge succeeds when the human stops repeating context: an agent remembers prior architectural decisions; issue clarification cites prior similar issues; a build plan references relevant receipts; a Ralph decision brief includes precedent; the platform recognizes that a repeated correction is a burden.

The crucial boundary discipline. GBrain is GLoops’ internal decision-history memory. It surfaces prior decisions as advisory precedent (influence: informed, never grounded); it cannot auto-ground actions or self-modify, and it is not a boundary surface — consumers read decision history by reading Receipts through the Capability API, never by reading GBrain directly. Authoritative grounding comes from Knowledge (knowledge://… URIs), not from memory of past decisions.

Five lenses — Knowledge

Lens
▣ Current Reality	Immutable receipts + the GBrain decision-history outbox drainer are built. But reasoning currently receives zero knowledge documents — wiring knowledge into reasoning is the build→trust step (`built (flagged)`).
◇ Target State	Organizational memory that compounds — ingesting documents, issues, PRs, receipts, ADRs, decisions, conversations, and code into a living substrate that grounds and informs every capability.
→ Migration Path	Ingest ADRs + CURRENT_STATE → ingest receipts → ingest issues/PRs → cross-link → context assembly for agents → user-correction memory.
⚖ Executive Implication	You explain the architecture once, not every session. Repeated corrections become capability improvements instead of recurring conversations.
☼ Operator Experience	Agents arrive already knowing the decision history; you stop being the organization’s memory.

Part III — The Operating Model

Chapter closeout

Decision ledger

Key Decisions

GBrain is advisory memory, never authoritative grounding or a boundary surface.

Open Questions

Which knowledge sources enter reasoning first without weakening provenance?

Related Chapters

Knowledge Graph Mockup
Receipt Explorer Mockup
Ch. 16 - Manual Rule