openclaw-memory-enhancer_skill

Overview

This skill provides an edge-optimized long-term memory system for OpenClaw that performs semantic search and automatic recall. It loads local memory files, builds lightweight vector indices, and supplies relevant context to conversations while keeping all data on-device. Designed for Jetson, Raspberry Pi, and other resource-constrained hardware (<10MB footprint).

How this skill works

The enhancer auto-loads files from a local memory directory and encodes content into compact vectors using a keyword + hash scheme (edge mode) or high-quality embeddings (standard mode). During a query it pre-filters by keywords, computes cosine similarity, ranks top matches, and returns formatted context for prompt augmentation. It stores memories locally in JSON or NumPy formats and maintains a memory graph to link related entries.

When to use it

• You need cross-session memory for an OpenClaw agent
• Running on edge devices with limited RAM/CPU (Jetson, Raspberry Pi)
• Building a lightweight knowledge base from chat history or logs
• Automatically surfacing relevant past interactions to improve responses
• Generating FAQs or summaries from accumulated conversations

Best practices

• Use the edge version on devices with <1GB RAM to keep memory under 10MB
• Organize memory files under memory/ with descriptive filenames and types (daily_log, preference, solution)
• Tune max_memory_size and similarity threshold for your workload to balance precision and recall
• Periodically export or compact memories to prevent fragmentation and stay under device limits
• Combine recalled context with concise prompts — include only top-k memories to avoid prompt bloat

Example use cases

• A customer-support OpenClaw bot that remembers user preferences and past tickets across sessions
• An embedded assistant on a Jetson device that recalls configuration steps and diagnostics
• Automatic FAQ generation for recurring user questions by consolidating qa memories
• Enhancing LLM prompts with context from recent conversations to reduce repeat questions
• Offline knowledge base for on-premise deployments where data must never leave the device