chaos-memory

What This Skill Does

CHAOS Memory (Context-aware Hierarchical Autonomous Observation System) is a robust, hybrid search memory framework designed for AI agents. Unlike standard context buffers that clear upon session termination, CHAOS provides persistent storage, allowing agents to maintain long-term memory across tasks. It uniquely utilizes four retrieval signals—BM25 for keyword matching, Vector embeddings for semantic understanding, Graph relationships for associative reasoning, and Heat tracking for access patterns—to ensure the most relevant information is retrieved. It is designed to be privacy-first, with all processing performed locally.

Installation

To integrate this memory system, ensure you have the OpenClaw environment ready. Use the following command to pull the skill into your workspace:

clawhub install openclaw/skills/skills/hargabyte/chaos-mind

Once installed, you can verify functionality by running chaos-cli --help. If you intend to use the advanced auto-capture features, ensure Ollama is installed locally and the qwen3:1.7b model is pulled to handle the intelligent extraction of insights from session transcripts.

Use Cases

1. Technical Project Management: Store architecture decisions, database connection strings, and research findings that need to be referenced weeks later across multiple coding sessions.
2. Context-Heavy Research: Synthesize findings from multiple sessions into a unified knowledge base, allowing the AI to query past experimental results without re-running computations.
3. Long-term Workflow Optimization: Use the Heat signal to automatically prioritize frequently accessed documentation or code snippets, reducing the time an agent spends searching through stale project data.

Example Prompts

1. "Check the CHAOS memory for any stored architecture decisions regarding our API gateway migration and summarize them in an index mode report."
2. "Store the fact that we have finalized the migration to Python 3.12, mark it with the category 'core', and set priority to 1.0."
3. "Search through all research-related memories from the last two weeks regarding model optimization and provide a summary of the most effective strategies."

Tips & Limitations

• Token Efficiency: Always leverage the --mode index flag when scanning for information to save up to 90% of your token budget. Reserve full mode only when detailed context is required for critical decision-making.
• Privacy: Auto-capture is explicitly disabled by default. If enabled, be mindful that it scans your local session transcripts. It is best practice to clean your local session directories periodically to avoid storing sensitive filler data.
• Maintenance: While the system handles relationships, occasionally using the chaos-cli list command helps you prune or update outdated entries to keep your memory signals accurate.