A secure, offline-first local embedding and semantic search system designed to provide deep contextual memory for Openclaw Skills on Windows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install node-llama-cpp-install-guide
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 node-llama-cpp-install-guide 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 Local Embedding Memory System for OpenClaw (Windows)?
The Local Embedding Memory System is a highly optimized, offline-capable memory enhancement suite built to run seamlessly on Windows environments. By leveraging lightweight local embedding models such as nomic-embed-text-v1.5 in GGUF format (~80MB), this system enables advanced semantic search capabilities for AI agents without relying on external cloud APIs. It forms a crucial performance backbone for Openclaw Skills, allowing local vector databases to cache, index, and retrieve historical context instantly.
In addition to local execution, this system addresses typical enterprise constraints, such as corporate firewalls and strict proxy configurations, by utilizing reliable mirror hosts (like hf-mirror) and supporting multi-tier configuration lockups to prevent runtime settings from being overwritten during gateway restarts. It also integrates an optional, beautiful 2D projection visualization utilizing UMAP and sentence-transformers, transforming raw memory vectors into actionable, interactive semantic maps.
Local Embedding Memory System for OpenClaw (Windows) Use Cases
- Offline Agent Context Retention: Running secure, high-performance semantic search for AI agents in intranet or air-gapped environments.
- Enterprise Security Compliance: Ensuring proprietary data used as agent memory never leaves the local network or Windows host.
- Interactive Memory Visualization: Generating interactive 2D HTML maps of agent memories using Nomic Atlas to debug memory storage and retrieval patterns.
- Intranet-Safe AI Pipelines: Bypassing strict firewalls by using local GGUF models or corporate-approved remote API proxies for embeddings.
How Local Embedding Memory System for OpenClaw (Windows) Works
- Local Model Ingestion: The system downloads and caches a highly quantized, local vector representation model (nomic-embed-text-v1.5) via robust mirrors into the user's local directory.
- Configuration Anchoring: The memory configuration is written and multi-tier locked across three levels (PRD presets, system environment variables, and whitelist files) to guarantee persistent settings across restarts.
- Plugin Activation: The
memory-tencentdbplugin (v0.3.8+) is installed to coordinate vector generation and full-text search (FTS) routines. - Contextual Retrieval: When executing Openclaw Skills, the agent queries the vector memory, generating high-precision semantic search results.
- Dimensionality Reduction & Visualization (Optional): Using Python (sentence-transformers & UMAP), high-dimensional embedding vectors are reduced to 2D coordinates and plotted onto an interactive HTML canvas for debugging.
Local Embedding Memory System for OpenClaw (Windows) Setup
Step 1: Download the Local GGUF Embedding Model
Run the following PowerShell command to create the directory and fetch the ~80MB nomic-embed-text model from a fast mirror host:
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.openclaw\models"
Invoke-WebRequest -Uri "https://hf-mirror.com/nomic-ai/nomic-embed-text-v1.5-GGUF/resolve/main/nomic-embed-text-v1.5.Q4_K_M.gguf" -OutFile "$env:USERPROFILE\.openclaw\models\nomic-embed-text-v1.5.Q4_K_M.gguf"
Step 2: Configure OpenClaw memorySearch
Add the following block to your openclaw.json configuration. Be sure to replace the modelPath user profile folder with your own Windows username:
{
"agents": {
"defaults": {
"memorySearch": {
"enabled": true,
"provider": "local",
"model": "fts-only",
"local": {
"modelPath": "D:\\Users\\YOUR_USER\\.openclaw\\models\\nomic-embed-text-v1.5.Q4_K_M.gguf"
}
}
}
}
}
Step 3: Install the Memory Plugin and Verify
clawhub install memory-tencentdb
openclaw memory status
Your expected status output should show:
Provider: local (requested: local)
Model: fts-only
Vector: ready
FTS: ready
Local Embedding Memory System for OpenClaw (Windows) Data Schema & Taxonomy
The local embedding memory skill uses a structured filesystem organization combined with local relational/vector files. High-dimensional vector metadata is organized as follows:
| Directory / File | Type | Description |
|---|---|---|
~/.openclaw/models/nomic-embed-text-v1.5.Q4_K_M.gguf |
Binary Model File | Core local embedding model for vector generation (~80MB). |
openclaw.json |
Configuration File | Agent-level settings mapping the local model path and enabling the FTS/vector providers. |
prd-openclaw.json |
Preset Configuration | The primary persistence level containing memorySearch configurations to avoid config loss on restart. |
nomic_atlas_visualizer.py |
Python Script | Generates interactive 2D HTML visual maps based on 768-dimensional sentence-transformer embeddings reduced via UMAP. |
Vector Metadata Taxonomy
- Dimension: 768 dimensions (Nomic Embed v1.5 standard).
- FTS Engine: Enabled concurrently alongside local vector space calculations.
- Quantization: Q4_K_M (4-bit quantization optimized for CPU execution on Windows).
Local Embedding Memory System for OpenClaw (Windows) Advanced Features
- Three-Layer Configuration Lock: Implements safety configurations to prevent SIGUSR1 signals or gateway reboots from wiping your setup. Protects configurations across
prd-openclaw.json, system environment variables viaopenclaw.cmd, and JS whitelist filtering. - Nomic Atlas 2D Vector Visualizations: Allows power users to export, project, and visualize spatial relationships of stored agent memories in interactive, browser-based 2D HTML maps using UMAP.
- Zero Native-Compilation GGUF Inference: Completely replaces fragile native module compilation dependencies (like node-llama-cpp) with pure GGUF memory indexing for foolproof Windows compatibility.
- Intranet Remote Failover: Built-in support to fall back from local GGUF models to highly secure enterprise proxy-compliant remote APIs (such as LiteLLM/OpenAI compatible gateways) when running in corporate air-gapped systems.
SKILL.md
Loading
Related Openclaw Skills
Technology Supply Chain Security Auditor
An advanced security auditor skill that maps tech supply chains, evaluates domestic substitution readiness, and simulates foreign trade sanction survival scenarios.
Groq Bot Skill
An ultra-fast, low-latency text generation skill utilizing Groq's free-tier models with automatic fallback mechanisms to handle rate limits seamlessly.
Make Change
An actionable toolkit and AI-driven framework for building social movements, organizing communities, and sustaining grassroots activism.
Supply Chain Volatility Watcher
An AI-powered intelligence agent designed to monitor raw material cost fluctuations, geopolitical trade disruptions, and upstream supply chain vulnerabilities.
im-still-here
An executable AI toolkit based on Austin Channing Brown's book to help users navigate white spaces, understand systemic exhaustion, and cultivate Black dignity and joy.
RDF / OWL Ontology Schema Drafting
Design semantic web and knowledge graph architectures by converting domain models into production-ready RDF and OWL ontology schemas.
Featured*
