Overkill Memory System

What This Skill Does

The Overkill Memory System (OMS) is a state-of-the-art memory architecture designed for OpenClaw agents, prioritizing millisecond-level latency while mimicking human neurobiology. By organizing information into six distinct brain-region-inspired tiers—including the Hippocampus for importance and the Amygdala for emotional tagging—the system ensures that the most relevant data is always accessible. It utilizes a hybrid approach, combining a fast-access vector database with Write-Ahead Logging (WAL) to prevent data loss. The system is engineered for a 5ms average query response time, making it significantly faster than standard retrieval-augmented generation methods. It features automatic self-reflection and error-learning, allowing the agent to refine its understanding based on user corrections. Additionally, the system incorporates spaced repetition via FSRS-6 to manage memory decay, ensuring long-term data retention without bloating the active query cache.

Installation

To integrate this memory system into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/broedkrummen/overkill-memory-system

Ensure that you have sufficient permissions for local file operations, as the system utilizes high-speed indexing tools like fd and rg for rapid local file searching.

Use Cases

• Complex Project Management: Maintain context across hundreds of sessions without losing track of specific project constraints.
• Agent Personalization: Allow your agent to recall personal preferences, emotional triggers, and historical interactions for a more human-like experience.
• Knowledge Retrieval: Rapidly query deep technical documentation or atomic facts stored within the integrated knowledge graph.
• Self-Improving Workflows: Automate diary entries and performance reviews to help the agent learn from previous mistakes and user feedback.

Example Prompts

• "Reflect on our last three meetings regarding the project architecture and tell me what potential errors we identified."
• "What is the most important emotional context I shared about my work goals last week?"
• "Find the atomic fact regarding the API endpoint security configuration stored in the knowledge graph."

Tips & Limitations

• Supermemory Usage: Keep in mind that cloud backups via Supermemory happen asynchronously. Do not expect immediate cloud-synced results during a live chat; rely on the local vector store for real-time needs.
• Performance: While the system is optimized for speed, extremely large datasets may impact the initial indexing time. Allow the background cron jobs to complete their maintenance cycles for optimal performance.
• Learning Curve: The system learns from your corrections. Be explicit in your feedback to the agent to maximize the efficacy of the error-learning protocol.