Cognitive Memory Blocks

What is a CMB

A Cognitive Memory Block is a set of typed field-vector pairs. Each field carries a symbolic text label (human-readable, for audit and retrieval) and a unit-normalised vector embedding (machine-comparable, for drift evaluation).

Every CMB also carries: source agent, origin timestamp (when the event happened), stored-at timestamp (when the CMB was created), confidence, and provenance (fusion history with per-field drift scores).

Example

Input: “user vibe coding with Claude for 8 hours straight, exhausted and losing focus, needs a break”

The fitness agent sees: mood (frustrated, low energy) is universally relevant (high affinity). Intent (needs a break) and issue (exhausted) are selectively relevant. Focus (debugging auth module) is domain-specific — the fitness agent ignores it. Per-field evaluation is impossible with a single flat embedding.

CAT7 — 7 Universal Fields

The 7 fields form a minimal, near-orthogonal basis spanning three axes of human communication. They are universal and immutable — domain-specific interpretation happens in the field text, not the field name.

Mood is the only fast-coupling field — affective state crosses all domain boundaries. The neural SVAF model confirmed this: mood emerged as the highest gate value (0.50) without supervision. All other fields couple at medium rates, with per-agent α_f weights controlling relative importance.

A coding agent’s focus is “debugging auth module.” A fitness agent’s focus is “30-minute HIIT workout.” Same field, different domain lens. The schema is fixed. The interpretation is sovereign.

Field-by-Field Guide

CMB vs. Raw Text

Per-Agent Field Weights

The schema is fixed. The weights are per-agent. New domains join the mesh by defining their weights — no schema changes, no protocol changes.

Regulated domains (legal, finance): issue and commitment always high — risks and obligations are non-negotiable.

Human-facing domains (music, fitness, health): mood always high — affect drives the experience.

Knowledge domains (coding, research): focus always high — subject matter is core.

Per-Agent Temporal Windows

The same signal has different temporal relevance depending on the receiver. Each agent defines its own freshness window.

How CMBs Fit the Stack

CMB flow is bidirectional per agent:

OUTBOUND: Agent observes → CMBEncoder (7 fields) → CMB → broadcast to mesh
INBOUND:  Mesh CMB arrives → SVAF evaluates per field → fused CMB

Each agent's xMesh LNN (Layer 6) processes BOTH flows:
    → own outbound CMBs (weighted by agent's α_f field weights)
    → inbound fused CMBs from mesh (SVAF-evaluated)
    → CfC evolves cognitive state (h1, h2)
    → state broadcast to peers (state-sync)
    → agent acts → remember() → new outbound CMB → loop

Training signal: CMB lineage.
    → CMBs that get remixed by other agents = valuable cognitive state
    → CMBs that expire without children = noise

CMBs are the data structure (what flows between agents). SVAF is the evaluation mechanism (Layer 4 — per-field drift and fusion).Synthetic Memory is the transformation layer (Layer 5 — CMB to CfC input). xMesh is the per-agent intelligence (Layer 6 — each agent’s own CfC, coupled via state exchange). Four distinct components, each independently defined.

Immutability & Lineage

A CMB is immutable after creation. It is never updated, never modified, never annotated. When an agent acts on a CMB, it creates a new CMB with lineage pointing back to the parent.

CMB-A (knowledge-feed): "Claude Cowork launched"
  └→ CMB-B (claude-code): "adjusted session to use Cowork"
       └→ CMB-C (melotune): "shifted to focus playlist"
            └→ CMB-D (melomove): "deferred exercise — user in flow"

Four agents, four CMBs, one lineage chain. No node writes to another node’s storage. Remix count is computed from the lineage graph, not stored on the parent. Like git commits. Like music remixes.

Each CMB carries two lineage fields: parents (direct — who I remixed) and ancestors (full ancestor chain — accumulated through every remix). Ancestors enable any agent in the chain to detect its CMB was remixed, even if it was offline during intermediate steps.

Ancestors also provide the endpoints for Mesh Cognition: when an inbound CMB arrives with ancestors, the agent’s LLM traces those endpoints to retrieve the remix subgraph and reasons about what happened and why — generating understanding that the agent’s previous state of mind didn’t have.

The mesh is a directed acyclic graph of immutable CMBs. Intelligence compounds through the graph — each cycle’s output becomes the next cycle’s input. No distributed state. No mutable counters. Just atoms and lineage.

Origin

Cognitive Memory Blocks were first formalised in the Mesh Memory Protocol (Consenix Labs, August 2025) with the CAT7 enterprise schema. The wellness/productivity schema and synthesis affinity classification were developed at SYM.BOT Ltd in late 2025 for production deployment across personal AI agents.