vector-memory-hack

What This Skill Does

The vector-memory-hack skill provides an ultra-lightweight, high-performance semantic search engine for your local OpenClaw memory files. By implementing a TF-IDF vectorizer coupled with SQLite storage, it enables your AI agent to bypass the token-heavy process of reading entire documentation files. Instead of loading thousands of tokens to find a specific rule or configuration, this skill computes sparse vectors to identify the most relevant document sections in under 10 milliseconds. It operates without external dependencies, making it a robust and portable solution for any environment where privacy and performance are paramount.

Installation

To install this skill, use the integrated clawhub installer. Run the following command in your terminal:

clawhub install openclaw/skills/skills/mig6671/vector-memory-hack

Once installed, initialize your database by running the rebuild command: python3 scripts/vector_search.py --rebuild. This will parse your existing MEMORY.md file and establish the local SQLite index needed for subsequent semantic queries.

Use Cases

• Context Loading: Automatically fetch relevant guidelines before executing complex system tasks.
• Large Documentation Handling: Efficiently query sprawling knowledge bases or project logs that exceed typical context window efficiency.
• Project Navigation: Quickly find specific procedural rules or API definitions buried in long markdown documents without wasting tokens on irrelevant text.
• Offline Operations: Perfect for air-gapped environments or scenarios where calling an external embedding API is prohibited or latency is an issue.

Example Prompts

1. "vsearch 'how do I handle the database migration process?'"
2. "Find the specific security guidelines for SSH access in my MEMORY.md file."
3. "Run a search for 'API authentication token requirements' to see if there are any existing rules I should follow."

Tips & Limitations

• Incremental Updates: Use the --update flag frequently if your MEMORY.md changes, as the index does not sync in real-time by default.
• Multilingual Support: The custom tokenizer is language-agnostic, but performance is optimized for languages where words are space-separated. For languages like Japanese or Chinese, you may need to adjust the tokenization logic in the script.
• Scope: This tool is designed for semantic similarity, not exact substring matching. If you need precise pattern matching, pair this with standard grep tools.
• Configuration: Ensure the MEMORY_PATH in vector_search.py points correctly to your primary documentation file to avoid search failures.