A zero-infrastructure, file-based long-term memory system for AI agents using BGE-M3 vector search and compound scoring.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install z1-memory-palace
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install z1-memory-palace using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Memory Palace?
Memory Palace is a sophisticated long-term memory architecture designed specifically for AI agents within the Openclaw Skills framework. Unlike traditional database-heavy solutions, it utilizes a zero-infrastructure, file-based approach that organizes knowledge into a logical palace structure. It empowers agents to maintain persistent context across sessions without the need for complex external services like Docker.
At its core, the system leverages BGE-M3 vector search to provide high-fidelity semantic retrieval. By implementing a multi-room directory structure—ranging from active chambers to the archive basement—Memory Palace ensures that agent knowledge is not just stored, but intelligently categorized and surfaced based on relevance, importance, and temporal decay.
Memory Palace Use Cases
- Building agentic long-term memory from scratch using Openclaw Skills standards.
- Upgrading flat-file memory systems with advanced vector search and temporal weighting.
- Managing agent knowledge through automatic consolidation and structured reflection.
- Implementing cold-storage archiving for large-scale agent datasets to maintain search performance.
- Establishing a GraphRAG Lite neighbor expansion for complex relational knowledge retrieval.
How Memory Palace Works
- The system initializes a structured directory hierarchy known as the palace to separate active knowledge from archived data.
- BGE-M3 embeddings are generated for all markdown files and stored in a local vector index via the build script.
- When a query is initiated, the system calculates a compound score based on semantic similarity (50%), recency (25%), and importance (25%).
- The GraphRAG Lite component expands the search by traversing internal links to provide contextual neighbor information.
- Memory metabolism protocols periodically promote high-value insights from project rooms to the reflection wing or constitution.
Memory Palace Setup
To get started with this component of Openclaw Skills, follow these steps:
# 1. Initialize the palace directory structure
mkdir -p palace/{grand_hall,chambers,project_rooms,reflection_wing,dispatch_corridor,conflict_room,archive_basement}
# 2. Install the necessary embedding dependencies
pip install FlagEmbedding
# 3. Build the initial vector index
python3 scripts/build_index_bge.py --force
# 4. Execute a test search query
python3 scripts/query_bge.py "your search query"
Memory Palace Data Schema & Taxonomy
The system uses a manifest-driven data schema to manage metadata and facilitate efficient filtering within the Openclaw Skills environment.
| Directory | Purpose |
|---|---|
grand_hall/ |
Global navigation, room maps, and system logs. |
chambers/ |
Agent-specific knowledge bases. |
reflection_wing/ |
Compiled insights, principles, and thinking patterns. |
archive_basement/ |
Cold storage excluded from default semantic search. |
Manifest Entry Format (watchdog_manifest_v1.jsonl):
path: The relative file path to the memory object.type: Metadata tag used for category-based filtering.priority: Importance level (high, medium, low) used in compound scoring calculations.
Memory Palace Advanced Features
- Compound Scoring Engine: Balances cosine similarity with 30-day half-life decay and manual priority weighting.
- Memory Metabolism: A 3-tier upgrade path that distills raw data into principles, prompt kernels, and failure patterns.
- Cold Zone Blinding: Specialized patches to prevent search noise by excluding the archive basement during active retrieval.
- Graph Router: Enables 1-hop link navigation to discover related knowledge nodes through internal file referencing.
- Incremental Maintenance: Efficiently updates the BGE-m3 vector index without rebuilding the entire database.
SKILL.md
Loading
Related Openclaw Skills
OpenClaw Three Manuals Generator
An interactive dialogue-driven tool designed to help users define their AI companion's identity and generate personalized SOUL.md, USER.md, and AGENTS.md configuration files.
Openclaw Heartbeat Designer
A specialized designer for establishing proactive AI agent inspection routines and automated heartbeat schedules to ensure consistent system monitoring.
OpenClaw Skill Compounder
A dynamic orchestration engine that chains multiple independent skills into automated super-skills for complex task execution.
Toutiao Article Details API
A specialized API tool for programmatically fetching detailed metadata and author information from Toutiao articles.
Stress Breathing Toolkit
A physiological menu of non-clinical breathing techniques and grounding exercises designed for immediate everyday stress management.
Blood Pressure Pattern Journal
A structured journaling framework designed to help users observe blood pressure patterns over time and prepare for clinical consultations.
Featured*
