Brain Cms

What This Skill Does

Brain CMS (Continuum Memory System) is a sophisticated architecture designed to evolve OpenClaw agents from basic file-reading assistants into entities with long-term, neuroscience-inspired memory. Rather than relying on simple, linear file parsing that consumes excessive token counts and fails on long-running projects, Brain CMS introduces a multi-layered approach. It acts as a cognitive engine using LanceDB and the nomic-embed-text embedding model to perform semantic retrieval. The system is comprised of several key components: a hippocampal router (INDEX.md) that directs information flow, an anchor system for high-significance event storage, and automated sleep cycles. During NREM phases, the system handles data compression and anchor promotion, while REM cycles utilize Ollama-powered consolidation to strengthen associations. By implementing this, your agent gains the ability to perform 'spreading activation,' only loading relevant schemas into context, which drastically improves both the quality of responses and the efficiency of your token budget.

Installation

Installation is streamlined for OpenClaw users via the clawhub CLI. First, ensure you have Python 3 and Ollama installed. Run the command: clawhub install openclaw/skills/skills/harrey401/brain-cms. Once installed, navigate to the workspace directory to initialize the virtual environment and install the required dependencies: cd ~/.openclaw/workspace/memory_brain && python3 -m venv .venv && .venv/bin/pip install lancedb numpy pyarrow requests. Finally, ensure your local Ollama instance has the necessary models pulled: ollama pull nomic-embed-text && ollama pull llama3.2:3b. After running the indexing script provided, the system will be ready to manage your agent's persistent memory architecture.

Use Cases

Brain CMS is best suited for complex, long-running agent tasks where context management is the primary bottleneck. Ideal use cases include: 1) Long-term software development projects where the agent needs to remember architecture decisions made weeks ago. 2) Research agents that aggregate massive amounts of literature and need to cross-reference data via semantic search. 3) Personal life management where the agent tracks habits, health data, and recurring project goals. 4) Enterprise agents that require domain-specific knowledge bases which are too large to fit into a standard prompt window.

Example Prompts

1. 'Brain, pull up the architectural decisions regarding the database schema from last month's meeting logs.'
2. 'Run a REM consolidation cycle for the current project to optimize the memory bank for upcoming tasks.'
3. 'Query the memory store for any references to the user authentication flow and update our INDEX.md mapping accordingly.'

Tips & Limitations

To maximize the performance of Brain CMS, maintain a clean INDEX.md. The quality of the 'Hippocampal router' dictates the agent's ability to locate information effectively. Avoid indexing trivial, ephemeral logs; instead, summarize them during the NREM phase. Note that this skill requires local compute resources for vector storage and embedding generation. If you find the agent is losing track, force an index rebuild using index_memory.py. Be mindful that while this reduces overall token usage by minimizing context stuffing, the initial embedding process does require significant processing time for large knowledge bases.