A high-performance local memory system for AI agents that integrates vector embeddings, fact storage, and knowledge graphs for comprehensive context recall.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-memory-tools
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 agent-memory-tools 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 Agent Memory Tools?
Agent Memory Tools is a robust framework designed to provide AI agents with persistent, reliable, and multi-layered memory. By utilizing Openclaw Skills, developers can implement a 100% local memory stack that eliminates cloud dependencies and API costs while maintaining high data privacy. The system leverages Ollama to run models like gemma3 locally, providing a seamless experience for developers building autonomous agents.
The core value of this skill lies in its hybrid retrieval architecture. It doesn't just rely on simple vector search; it combines a structured fact store, vector embeddings, BM25 full-text search, and an automatically generated knowledge graph. This multi-source approach ensures that Openclaw Skills can recall precise facts, semantic concepts, and complex relationships across vast workspaces with high accuracy and minimal latency.
Agent Memory Tools Use Cases
- Retrieving precise workspace knowledge across diverse documentation and codebases.
- Automatically extracting and categorizing atomic facts from raw text or conversation logs.
- Maintaining data integrity by detecting contradictions in newly ingested information.
- Building complex entity-relationship graphs to enable multi-hop reasoning for AI agents.
- Implementing temporal memory where recent facts are prioritized over stale data using decay scoring.
How Agent Memory Tools Works
- Unified Querying: The system receives a natural language query and triggers a unified recall process.
- Multi-Source Retrieval: It simultaneously queries the local fact store, vector database, BM25 index, and knowledge graph.
- Scoring and Reranking: Results from all four sources are merged, scored based on relevance and recency, and reranked.
- Context Synthesis: The most relevant information is passed to a local LLM to generate a synthesized answer.
- Automated Ingestion: A background watcher monitors file changes, extracts new facts, checks for contradictions, and updates the knowledge graph incrementally.
Agent Memory Tools Setup
To deploy these Openclaw Skills, you need Python 3.9+ and Ollama installed on your local machine. Follow these steps to initialize the environment:
# Install required local models
ollama pull gemma3:4b
ollama pull nomic-embed-text-v2-moe
# Run the self-validation script
python3 scripts/selftest.py
Configuration can be customized in scripts/config.json. The system automatically detects if you are using a local JSON store or a Convex backend for shared memory.
Agent Memory Tools Data Schema & Taxonomy
The skill manages its data through a structured hierarchy in the .cache directory, allowing for efficient local retrieval and indexing.
| Component | Storage Type | Taxonomy / Categories |
|---|---|---|
| Fact Store | JSON / Convex | knowledge, error, timeline, preference, tool, client, hr |
| Embeddings | Vector Cache | Nomic-based semantic vectors |
| Knowledge Graph | JSON | Entity-relationship mappings and triples |
| Ingestion Tracker | Hash Map | Content hashes and cooldown timers for file syncing |
Agent Memory Tools Advanced Features
- Temporal decay search that adjusts relevance based on a configurable half-life (e.g., 14 days).
- Multi-hop search capabilities that chain multiple LLM calls to navigate the knowledge graph.
- Native contradiction detection that prevents the agent from storing conflicting information.
- Platform-specific automation support for macOS (LaunchAgents), Linux (systemd), and Windows (Task Scheduler).
- Flexible model presets allowing instant switching between Ollama, LM Studio, or OpenAI backends.
SKILL.md
Loading
Related Openclaw Skills
EmailOctopus Skill
A powerful integration for managing EmailOctopus lists, campaigns, and templates through AI-driven automation and secure credential management.
NotebookLM PDF Cleaner
A precision tool designed to mask the NotebookLM footer badge on exported slide-decks for a professional, presentation-ready finish.
Memory Manager
A sophisticated dual-track memory system that prevents AI context loss through automated daily journaling and long-term knowledge distillation.
XLSX Spreadsheet Master
A professional-grade skill for automating complex Excel workflows, including formula recalculation, financial modeling, and data visualization.
Garmin Connect Data Integration
A specialized skill for extracting and analyzing health, sleep, and activity metrics directly from Garmin Connect wearable devices.
WeryAI Video Tool Magic Style
A specialized AI skill for applying professional visual style transfers to existing videos via public URLs.
Featured*
