tiered-memory

What This Skill Does

The tiered-memory skill is an advanced cognitive framework for the OpenClaw AI agent, modeled after human memory structures and PageIndex tree retrieval. It manages information across three distinct tiers—Hot, Warm, and Cold—to ensure the agent remains context-aware without overwhelming its active memory window.

In version 2.2.0, the skill introduces automated daily note ingestion. This feature bridges the gap between daily journals and long-term memory by automatically processing files from memory/YYYY-MM-DD.md into the tiered system. The architecture handles structured metadata extraction, specifically preserving URLs, shell commands, and file paths. By performing ongoing distillation and consolidation, the system ensures that vital information persists through the 10-year archival tier, while transient, less critical facts decay from the 30-day warm memory buffer.

Installation

To integrate this memory system into your agent, use the official clawhub installer. Ensure your environment has write access to your local memory directory.

clawhub install openclaw/skills/skills/bowen31337/tiered-memory

Use Cases

• Personal Knowledge Management: Automatically tracking project progress, meeting notes, and daily insights without manual entry.
• Technical Documentation Tracking: Preserving shell commands, configuration snippets, and project-specific URLs so they are instantly searchable during development.
• Long-term Project Continuity: Retaining lessons learned from tasks performed weeks or months ago, allowing the agent to provide better guidance based on past outcomes.
• Information Synthesis: Bridging disconnected data paths by automatically consolidating daily entries into a structured, queryable knowledge graph.

Example Prompts

1. "Check my recent notes and remind me of the specific URL I saved for the production API docs last week."
2. "Review the last 30 days of memory. Are there any incomplete tasks or missing commands that I need to address regarding the current project?"
3. "Summarize the key technical decisions made regarding the database migration from my recent daily notes."

Tips & Limitations

• Optimization: Keep your daily notes clean and well-structured; the LLM's metadata extraction is robust, but clear formatting helps significantly.
• Cold Storage: Understand that the 'Cold' tier (Turso DB) is optimized for long-term retention. Retrieval from this tier may have slightly higher latency than local Hot/Warm memory.
• Active Context: Do not bloat your 'Hot' memory with unnecessary facts. Use the system's inherent distillation, or let the skill handle decay, to prevent prompt degradation.