Use Core Memory
Core Memory helps AI agents stay consistent over time by preserving not just what happened, but why it happened.
- Deterministic causal recall: stable tie-breaks and explainable results with replay hashes.
- Two-factor promotion: score plus reinforcement, with adaptive thresholds to prevent memory inflation.
- Session-start continuity: token-budgeted rolling-window context selected deterministically every restart.
pip install core-memory
FAQ
Can Core Memory explain exactly why a memory result was chosen?
Yes. Core Memory uses deterministic causal recall with explicit tie-break policies down to bead_id. Explain reports can include replay hashes, and ordering stays stable across restarts.
How does Core Memory avoid memory inflation over long runs?
Core Memory uses two-factor promotion: score plus reinforcement must both pass. The threshold then self-raises as the promoted ratio grows.
How does Core Memory keep continuity at the start of a new session?
Core Memory injects a token-budgeted rolling window from distilled historical beads at session start. Selection is deterministic and idempotent, a pattern not typically used by Mem0 or Zep.