learn

To teach your AI agent to remember preferences across sessions, use a memory-writing workflow that captures behavioral corrections and facts, storing them immediately in a markdown vault for future recall.

Conflict resolution in an agent memory vault works by grepping for overlapping active or needs-review topics, then supporting update, supersede, or separate resolution to prevent contradictory behavioral preferences.

The best way to store persistent facts for agent learning without data loss is creating one memory file per concept, ensuring each distinct fact is stored independently to support independent deprecation later.

You can track changes and corrections when saving behavioral preferences to a knowledge base by formatting audit-log entries during the memory indexing updates, providing a clear record of all modifications.

When creating one file per concept in a memory vault, filename collision handling is applied during the memory-writing workflow to ensure each distinct fact receives a unique, accessible file path.