Research

Synthetic Memory

The Transformation Layer for Collective Intelligence

Author: Hongwei XuVersion 0.1.0March 2026SYM.BOT Ltd

Core Insight

Memory is not shared. Memory is not copied. Memory is synthesized. When Agent A broadcasts an observation and Agent B receives it, Agent B does not store Agent A’s memory. Agent B synthesizes a new memory — one shaped by its own domain intelligence. SYM is the mechanism that makes this possible.

What SYM Does

SYM transforms any agent’s memory into input for xMesh — the meshed Liquid Neural Network where collective intelligence emerges. The agent does not need its own neural network. SYM is the transformation layer that enables memory synthesis.

Agent Memory → SYM (transform) → xMesh (coupled CfC) → Collective Intelligence → Agents

1TRANSFORM

SYM encodes agent synthetic memory (insights from mesh signals) into CfC-compatible hidden state vectors

2FEED

Encoded vectors become input to xMesh — the meshed LNN

3RETURN

Collective state flows back to agents as synthesized context

Synthetic ≠ Shared

Shared Memory	Synthetic Memory
Copied between agents	Each agent synthesizes new insight from mesh signals
Same data everywhere	Different synthesis per domain
Like a shared database	Like how humans form memories from conversation
Central store needed	Fully decentralised — each agent owns its memory
Data sharing	Intelligence emergence

Key distinction Synthetic memory is not what other agents observed. It is insight that emerges when an agent processes mesh memory through its own domain intelligence. The agent doesn’t store what others saw. It stores what it understood.

What Synthetic Memory Produces

Agents receive the same mesh signals. Each agent synthesizes different insight — shaped by its own domain intelligence. The output is not a copy of what other agents observed. It is new understanding that only that agent could produce.

Synthetic memory feeds forward into xMesh, where coupled CfC dynamics produce collective intelligence. The understanding compounds over time.

The SYM Encoder

SYM: M(agent) → (h₁, h₂) ∈ ℝᵈ × ℝᵈ

The SYM Encoder converts any agent’s memory — natural language, structured data, sensor readings, and synthesized insights — into a CfC-compatible hidden state vector pair. This is the input format for xMesh’s coupled dynamics.

Universal

Accepts any text content. Agents contribute memory in their own format.

Semantic

Similar content produces similar vectors. Preserves meaning.

CfC-compatible

Output matches CfC hidden state format (h₁, h₂ pair).

Lossy

Compressed representation. Forces abstraction — detail becomes pattern.

Where SYM Sits

SYM is Layer 5 in the Mesh Memory Protocol stack. Together with Layer 4 (Coupling) and Layer 6 (xMesh), it forms Mesh Cognition.

Layer 7  APPLICATION     Domain Agents
─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
Layer 6  xMesh           Meshed LNN            ┐
Layer 5  SYM             Synthetic Memory      ├ Mesh Cognition
Layer 4  COUPLING        Drift · SVAF          ┘
─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
Layer 3  MEMORY          L0 · L1 · L2
Layer 2  CONNECTION      Handshake · Gossip
Layer 1  TRANSPORT       IPC · TCP · WS · Push
Layer 0  IDENTITY        nodeId · keypair

Layers 0–3 Identity, transport, connection, and memory storage.Layer 6 xMesh runs coupled CfC dynamics on SYM-transformed inputs. Intelligence emerges there.

How Synthesis Works

Claude Code → SYM → "debugging 3hrs, stressed" → h_claude
MeloTune   → SYM → "replaying ambient tracks"   → h_tune
MeloMove   → SYM → "sedentary, no activity"      → h_move

        ┌─────────────────────────────────────┐
        │  xMesh (coupled CfC dynamics)       │
        │                                     │
        │  Fast neurons (τ small): emotion    │
        │  → synchronize across agents        │
        │                                     │
        │  Slow neurons (τ large): expertise  │
        │  → stay sovereign per agent         │
        │                                     │
        │  Output: collective state           │
        │  → no single agent had this view    │
        └─────────────────────────────────────┘

Collective state flows back:
  → MeloTune synthesizes: play stress-relief music
  → MeloMove synthesizes: generate recovery workout
  → Claude Code synthesizes: "a break helped last time"

No agent was told what to do. Each agent received the collective state from xMesh and synthesized its own response through its domain intelligence. SYM made the input possible. xMesh made the intelligence emerge.

Install

npm install @sym-bot/sym

Three lines to give your agent synthetic memory. SYM transforms its synthetic memory into xMesh input. The coupling engine evaluates relevance autonomously. No routing rules. No configuration.

Mesh Memory Protocol (MMP) — the protocol stack that includes SYM

Mesh Cognition Whitepaper — CfC neural networks, Kuramoto coherence, coupling theory

Intellectual Property

Synthetic Memory (SYM) is original work by Hongwei Xu and SYM.BOT Ltd. The following remain proprietary: the SYM transformation mechanism, synthesis processes, trained encoders, xMesh coupled dynamics, domain-specific product integrations, and production configurations.

Academic citation of this work is permitted and encouraged.

For partnership inquiries: info@sym.bot