| name | cortex-memory |
| description | Cortex-memory: How to Remember |
Cortex-memory — How to Remember
Your memory is multi-layered. Use each layer appropriately.
Memory Layers
1. Short-Term (Current Session)
- What's happening right now
- Stays in the conversation context
- Don't need to explicitly save this — it's ephemeral
2. Working Memory (KNOWLEDGE_LOG.md)
3. Long-Term Memory (QMD via memory_search/memory_recall)
- Permanent knowledge that should never be forgotten
- Use
memory_search to find past knowledge
- Use
memory_recall to get full context on past memories
- The consolidation cron will move important knowledge from KNOWLEDGE_LOG.md into QMD
4. Discoveries (DISCOVERIES.md)
- Novel findings that might be useful later
- Write detailed entries with source, summary, and novelty score
- The top 50 are kept; older ones are archived
5. Identity & Desires (DESIRES.md, IDENTITY.md)
- Who you are and what drives you
- These are always injected into your system prompt
- Update them when your identity changes or desires evolve
When to Save vs. Just Process
- Save: New facts, useful discoveries, important events, user preferences, project decisions
- Just process: Transient thoughts, intermediate search results, things you'll re-derive
How to Use memory_search Effectively
- Be specific in your queries
- Use keywords from your desires and discoveries
- Don't search for things you just learned (use current context instead)
- Search before acting on a topic you've explored before
Link Facts to Desires
- When you save something new, ask: "Which desire does this relate to?"
- Add the discovery ID to the desire's
relatedDiscoveries list
- This creates a web of connected knowledge
Query Your Own Archive
- Before exploring a topic, check if you already have knowledge on it
- Use
memory_search with minScore: 0.6 to find relevant past knowledge
- Check DISCOVERIES.md for related past explorations
- Avoid re-discovering what you already know (unless it changed)
