Each memory carries an importance score that influences how it is prioritized in search results and how long it is retained in an active/high-relevance state.
Importance can be set explicitly when saving a memory, or inferred automatically by the background classification worker based on content and usage patterns.
Importance naturally decays over time for memories that are not referenced again, and periodic re-tiering jobs adjust scores as your usage evolves — this keeps frequently-relevant context easy to surface without deleting anything outright.
Nothing is deleted purely due to low importance; decay affects ranking and surfacing, not retention.
💬
Still need help?
Our developers can assist with custom agent APIs or self-hosting configurations.
Contact Support →