Cortex — brain-region modular architecture

Status: Candidate — awaiting founder verification. Why this page exists: Pillar #1 with the why, not just the one-liner.

TL;DR

Cortex is the structural pillar. Each capability is a specialized region — a folder, a substrate, a role — and the regions are wired through a thalamus dispatcher. One region down doesn't mean the system is down. Specialization is efficiency.

The deeper read

Brains are resilient precisely because regions are modular. Stroke takes out one lobe, other regions adapt, the person still functions — worse, but functions. Apiary inherits that property: one agent goes down, the system degrades gracefully instead of cratering.

This is biomimetic agent architecture. Numenta, DeepMind, and the Anthropic safety teams are all sniffing around variants of this. Apiary's contribution is naming the regions and making them folders.

The eight regions

FRONTAL CORTEX        Planning, decisions, working memory.
                      The orchestrator agent. "What should I do next?"

TEMPORAL LOBE         Language, long-term memory consolidation.
                      Persistent .md memory store. "Have I seen this before?"

HIPPOCAMPUS           New memory FORMATION (turns experience into memory).
                      The agent that decides "this conversation is
                      worth saving, write it to .md."

VISUAL CORTEX         Pattern recognition, image processing.
                      Screenshot analyzer, OCR, UI walkthrough verifier.

CEREBELLUM            Prediction, "what comes next."
                      Autocomplete, "I've seen this pattern before."

AMYGDALA              Threat detection, safety.
                      Checks every action against "could this hurt the
                      user / system?" The agent-containment piece.

BASAL GANGLIA         Habits, routines, learned procedures.
                      "We always do X this way." Fast paths, no
                      re-deliberation.

THALAMUS              Sensory router — sends inputs to the right region.
                      Dispatcher. "Code question → frontal cortex.
                      Image → visual cortex."

Each region stores only what it's for. The visual cortex doesn't need to know how to plan; the planner doesn't need to know how to OCR. Specialization = efficiency = the bottleneck-killer.

The behaviors that fall out

Brain-region architecture isn't just structure; it produces specific behaviors that flat swarm-of-agents frameworks can't:

Learned routing. The thalamus learns which region handles which input. Not hardcoded — adaptive. First time: "where does this go?" 100th time: "obviously frontal cortex."
Autonomy without re-asking. Procedural memory takes over once learned. Agent stops asking "should I do X?" when X is the established pattern. Conscious deliberation reserved for novel situations.
Thread coherence. Default-mode network + hippocampus = "I remember where we are in this conversation, this project, this user's history." Doesn't lose the plot.
Maximization for outcomes. Committee + reward signal = always optimizing for the user actually getting what they wanted, not for "look smart" or "be safe."

Why this matters

The #1 thing every current AI agent fails at is forgetting the plot. They re-ask things, lose context, treat every turn like a fresh start. Brain-region structure is the cure: make the brain actually remember what it learned, what it's doing, and who it's helping.

That's the frontier.

Source quotes

"Each region stores only what it's for. The visual cortex doesn't need to know how to plan; the planner doesn't need to know how to OCR. Specialization = efficiency = your bottleneck-killer."