Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Replaces flat MEMORY.md files with a multi-layer memory architecture for OpenClaw agents. Uses a hippocampal router, LanceDB vector store, and automated NREM/REM sleep cycles to load only relevant context per session. Modeled on neuroscience concepts like long-term potentiation and spreading activation.
Loads 50-line context cores with on-demand schema retrieval instead of injecting 150-300 lines every session, cutting token usage by 40-60%.
When to use it
- Reducing token costs on long-running agents
- Setting up persistent memory across agent sessions
- Semantic search over accumulated agent knowledge
- Auto-consolidating daily logs into structured schemas
- Preserving high-significance events as permanent anchors
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: 'What happened in last week's project demo?'
- 1Check memory/INDEX.md for relevant topic triggers
- 2Load matching semantic schema from memory/<topic>.md
- 3Query ANCHORS.md for any CRITICAL-tagged events
- 4Run vector similarity search via query_memory.py if topic is ambiguous
- 5Return consolidated context from matched layers
Agent retrieves only the relevant episodic and anchor entries instead of loading the full memory file
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.