A temporal Graph-RAG system providing AI agents with persistent, queryable long-term memory through a Mixture-of-Experts knowledge graph architecture.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install graph-rag-memory
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 graph-rag-memory 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 Graph-RAG Memory Skill?
The Graph-RAG Memory skill provides a sophisticated long-term memory layer for AI agents, leveraging the Graphiti framework and FalkorDB. It enables agents to transcend the limitations of context windows by persisting facts, entities, and relationships within a temporal knowledge graph. This architecture ensures that information is not just stored but remains contextually relevant as relationships evolve over time.
By integrating this within the Openclaw Skills ecosystem, developers can utilize a Mixture-of-Experts (MoE) router to handle complex data ingestion. The system automatically categorizes information into domains such as technical, personal, or research, utilizing specialized embedding models to ensure high-precision retrieval during agent operations.
Graph-RAG Memory Skill Use Cases
- Creating a persistent memory bank for AI agents to remember user preferences and past project decisions across multiple sessions.
- Ingesting large-scale technical documentation and codebases into a structured knowledge graph for fast, relational querying.
- Building research assistants that track the temporal evolution of facts and entity relationships.
- Enhancing RAG pipelines with hybrid search that combines keyword-based BM25 and semantic vector similarity.
- Managing cross-agent knowledge sharing through a centralized FalkorDB graph.
How Graph-RAG Memory Skill Works
- Content Processing: Text content from conversations or documents is passed through a DomainRouter.
- Expert Embedding: The router selects a domain-specific embedder (e.g., technical or general) based on centroid routing or metadata.
- Entity Extraction: An LLM identifies entities and relationships, which are then formatted as episodes for the temporal graph.
- Graph Storage: Data is persisted in FalkorDB, where nodes represent entities and edges represent time-aware relationships.
- Hybrid Retrieval: When a query is made, the system performs a parallel search using both vector embeddings and BM25 indexing.
- Ranking: Results are fused using Reciprocal Rank Fusion (RRF) to provide the agent with the most relevant factual context.
Graph-RAG Memory Skill Setup
To deploy this skill, ensure FalkorDB and Ollama services are accessible within your environment. Follow these steps for installation:
# Install the necessary Python dependencies
pip3 install --user --break-system-packages graphiti-core falkordb sentence-transformers
# Configure the service endpoints in memory-upgrade/config.py
# Example: OLLAMA_URL = "http://172.18.0.1:11436"
# Initialize the memory graph and run initial ingestion
python3 memory-upgrade/phase3_ingest.py
# Verify the system status
python3 memory-upgrade/scripts/status.py
Graph-RAG Memory Skill Data Schema & Taxonomy
The skill organizes data within a specialized graph structure to facilitate rapid retrieval and temporal accuracy. Data is indexed as follows:
| Component | Type | Description |
|---|---|---|
| Entities | Nodes | Typed nodes representing people, projects, technologies, or concepts. |
| Relationships | Edges | RELATES_TO connections containing temporal metadata and fact descriptions. |
| Vector Index | HNSW | 768-dimensional cosine similarity index for semantic search. |
| Checkpoints | JSON | State files stored in /checkpoints to track ingestion progress and prevent data duplication. |
| Domains | Metadata | Categorization (e.g., personal, technical) used for MoE routing. |
Graph-RAG Memory Skill Advanced Features
- Mixture-of-Experts (MoE) routing architecture for specialized domain embedding.
- Temporal graph capabilities allowing agents to query facts at specific points in time.
- Hybrid search fusion combining BM25 and vector similarity for superior retrieval accuracy.
- Checkpoint-aware ingestion scripts designed for high-volume workspace data processing.
- Centroid-based routing for automatic selection of the best embedding model for a given query.
- Integration hooks for Openclaw Skills to support multi-agent memory sharing.
SKILL.md
Loading
Related Openclaw Skills
PDF Translator
A sophisticated translation engine that preserves complex document layouts by leveraging LaTeX reconstruction and direct arXiv source integration.
Mind-Wander
An autonomous background reasoning agent that explores open questions and hypotheses using local models without consuming your primary token budget.
Hyperliquid Names (HLN) API
An AI agent skill designed to interact with the Hyperliquid Names API for domain resolution, profile lookups, and HyperEVM integration.
Awesome Closet Stylist
A high-autonomy AI agent skill that transforms your digital wardrobe into a personal stylist through natural language interaction and intelligent outfit planning.
SQL Database Toolkit
A comprehensive data analysis suite that integrates database connectivity, SQL execution, interactive visualization, and AI-powered report generation.
OpenClaw Pilot
OpenClaw Pilot is an instruction-first skill that compiles rough ideas into a human-readable blueprint and a clean, machine-ready OpenClaw execution packet.
Featured*
