Everyone builds agent memory for store-and-search. I built the other half: write and delete.

Description and use

memgovern adds write, delete, and conflict handling to agent memory, enabling decay, reversible deletions, and manual conflict resolution.

Problem

Agent memory systems focus on storing and retrieving embeddings but ignore how memories should be written, deleted, and reconciled when conflicting.

Outcome

memgovern adds write, delete, and conflict handling to agent memory, enabling decay, reversible deletions, and manual conflict resolution.

Workflow

• Initialize MemoryStore with conflict_policy=MANUAL
• Write a memory with importance
• Write a conflicting memory on same key
• Resolve conflict by choosing winner
• Delete a memory with a reason
• Audit a key to view history

Evidence

memgovern treats these as first-class operations. Zero dependencies, SQLite under the hood:

Source

Decay: memories carry importance + TTL; queries rank by exponential-decay score, not recency alone.

Source

Tombstones: deletes are tombstones, not erasures — reversible, with a reason attached, full audit trail.

Source

Conflict arbitration: a contradictory write on the same key is quarantined as PENDING. You pick the winner (new / old / both / human decides) instead of silently clobbering.

Source