openclaw-memories

What This Skill Does

The openclaw-memories skill provides a robust, tripartite architecture for managing long-term memory in AI agents. By integrating ALMA (Algorithm Learning via Meta-learning Agents), an intelligent Observer, and a local Indexer, it ensures your agent can recall past interactions, evolve its own memory configuration, and query local knowledge bases effectively.

Core Components:

• ALMA: This component handles memory optimization. Using gaussian mutation and simulated annealing, ALMA autonomously proposes and tests different memory structures, keeping only the top-performing designs to ensure your agent remains efficient over time.
• Observer: Acting as the agent's cognition layer, the Observer leverages LLM APIs (OpenAI, Anthropic, or Gemini) to parse raw conversation logs. It extracts structured entities, priorities, and confidence scores, allowing the agent to distinguish between subjective opinions and hard world facts.
• Indexer: A high-performance local tool that scans your workspace Markdown files (daily logs, entity banks, and memory files). It enables full-text search, allowing the agent to retrieve relevant context from your existing documentation without needing network access.

Installation

To add this skill to your OpenClaw agent, execute the following command in your terminal:

clawhub install openclaw/skills/skills/arosstale/openclaw-memories

Alternatively, for manual project integration via npm:

npm install @artale/openclaw-memory

Use Cases

• Long-term Project Management: Maintain a persistent history of project decisions and requirements across multiple sessions by keeping daily logs in the memory folder.
• Personalized AI Persona: Use the opinion tracking features to ensure the agent remembers your preferences, confidence levels in certain topics, and historical views.
• Knowledge Retrieval: Use the Indexer to point the agent at a folder of technical documentation or notes, enabling it to search and answer questions based on your specific Markdown content.

Example Prompts

1. "Search my daily logs from last week to find where we left off with the API integration project."
2. "Extract the key facts from our conversation and update the memory files with the new entity definitions."
3. "Run ALMA to optimize the current memory design and show me the top 5 ranking configurations."

Tips & Limitations

• API Costs: The Observer requires an active API key for OpenAI, Anthropic, or Gemini. If no key is provided, the Observer component will fail silently.
• Local-First: ALMA and the Indexer operate entirely offline, providing excellent privacy and zero-latency performance for your internal document management.
• Mock DB: Note that the Indexer currently uses an in-memory mock database. It is highly efficient for smaller project workspaces but lacks the advanced features of a production-grade SQLite FTS5 implementation.
• Network: While most of the system is offline, the Observer functionality strictly requires network access to call remote LLM endpoints.