DEV Community

Aloysius Chan
Aloysius Chan

Posted on • Originally published at insightginie.com

Brain CMS: A Neuroscience-Inspired Memory System for OpenClaw Agents

#news #insights #ginie #openclaw

What Is Brain CMS?

Brain CMS (Continuum Memory System) is a neuroscience-inspired memory
architecture for OpenClaw agents that replaces traditional flat file injection
with a sophisticated multi-layer memory system. This approach dramatically
improves context efficiency while reducing token costs for long-running
agents.

Core Architecture

The system organizes memory into distinct layers based on how human brains
store and retrieve information:

  • Working Memory : The lean core (MEMORY.md) plus today's daily log, loaded every session
  • Episodic Memory : Daily logs stored as memory/YYYY-MM-DD.md, loaded during boot
  • Semantic Memory : Domain-specific schemas loaded on trigger
  • Anchors : High-significance events in memory/ANCHORS.md, loaded for critical topics
  • Vector Store : LanceDB-powered semantic search for ambiguous queries

Key Components

The Brain CMS installation creates a structured directory system:

memory/
├── INDEX.md          # Hippocampus: topic router + cross-links
├── ANCHORS.md        # Permanent high-significance event store
└── schemas/          # Domain-specific semantic schemas
memory_brain/
├── index_memory.py   # Embeds schemas into LanceDB vector store
├── query_memory.py   # Semantic similarity search
├── nrem.py           # NREM sleep cycle (compression + anchor promotion)
├── rem.py            # REM sleep cycle (LLM consolidation via Ollama)
└── vectorstore/      # LanceDB database (auto-created)
Enter fullscreen mode Exit fullscreen mode

How It Works

The system follows a sophisticated retrieval pattern:

  1. Boot Sequence : Load MEMORY.md (lean core) + today's daily log
  2. Topic Detection : When a topic appears, read memory/INDEX.md to find relevant schemas
  3. Spreading Activation : Load only the schemas related to the detected topic
  4. : Check memory/ANCHORS.md for high-significance events
  5. Ambiguous Topics : Run semantic search using the vector store

Automated Sleep Cycles

The system mimics biological sleep processes:

NREM Sleep Cycle

Run on shutdown (approximately 30 seconds, no LLM required):

cd ~/.openclaw/workspace/memory_brain && .venv/bin/python3 nrem.py
Enter fullscreen mode Exit fullscreen mode

This compresses memories and promotes anchors to permanent storage.

REM Sleep Cycle

Run weekly (2-5 minutes, uses local llama3.2:3b model):

cd ~/.openclaw/workspace/memory_brain && .venv/bin/python3 rem.py
Enter fullscreen mode Exit fullscreen mode

This performs LLM-based consolidation of memories.

Setup and Installation

Setting up Brain CMS requires a one-time installation process:

  1. Run the installer : python3 ~/.openclaw/workspace/skills/brain-cms/install.py
  2. Index your schemas : cd ~/.openclaw/workspace/memory_brain && .venv/bin/python3 index_memory.py
  3. Test retrieval : .venv/bin/python3 query_memory.py "your topic here" --sources-only

Semantic Schemas

When new significant projects or domains appear, create memory/.md
files and add them to INDEX.md with triggers, priority, and cross-links. The
system automatically re-indexes when schemas change.

Performance Benefits

Brain CMS offers substantial performance improvements:

  • Typical MEMORY.md : 150-300 lines injected every session
  • With Brain CMS : ~50-line core + schemas loaded only when relevant
  • Estimated savings : 40-60% reduction in context tokens per session

Technical Requirements

The system requires:

  • Python 3.10+
  • Ollama (for embeddings + REM consolidation)
  • 500MB+ storage for vector store and models
  • Python packages: lancedb, numpy, pyarrow, requests (auto-installed)

Tagging Anchors

In daily logs, tag high-significance events using the [ANCHOR] tag:

[ANCHOR] Major demo success — full pipeline working end-to-end
Enter fullscreen mode Exit fullscreen mode

The NREM cycle automatically promotes these to ANCHORS.md on next shutdown.

Why Choose Brain CMS?

Brain CMS is ideal when you need:

  • Persistent agent memory across sessions
  • Improved context efficiency for long-running agents
  • Reduced token costs through selective memory loading
  • Semantic search capabilities for ambiguous queries
  • A neuroscience-based approach to memory management

By implementing Brain CMS, you transform your agent's memory from a simple
file dump into a sophisticated, efficient, and intelligent system that mimics
how biological brains actually process and store information.

Skill can be found at:
cms/SKILL.md>

Top comments (0)