Skip to main content
Mneno preserves memory history instead of silently deleting or overwriting it. Conflict resolution and hierarchy transitions are deterministic, local, and recorded in each memory’s audit trail.

​
Memory statuses

Memories use four lifecycle statuses:
  • active: available for normal retrieval
  • superseded: replaced by a newer memory but retained for history
  • archived: retained but excluded from default retrieval
  • conflicted: linked to another memory with an unresolved contradiction
Search and context building exclude archived and superseded memories by default. Use explicit inactive-inclusion options only when you need historical records.

​
Detect conflicts while adding

Conflict detection runs automatically by default. Use add_with_report() when you need the reports and resolution actions. 
from mneno import MemoryClient

client = MemoryClient()
client.add("User prefers Python 3.10.", memory_type="preference")

result = client.add_with_report(
    "User now prefers Python 3.11.",
    memory_type="preference",
)

for report in result.conflict_reports:
    print(report.conflict_type, report.reason, report.suggested_action)

print(result.resolution_actions)
Mneno never deletes memories during conflict resolution. Supersession links the old memory through superseded_by, while contradictions use conflicts_with links.

​
Detect without mutating storage

reports = client.detect_conflicts("User does not prefer Python.")
Use client.list_conflicts() to reconstruct stored conflict reports from audit history.

​
Memory layers

Mneno organizes memories into these layers:
  • short_term
  • working
  • episodic
  • semantic
  • operational
  • archived
Operational, working, and semantic memories receive higher default retrieval priority. Retention decisions use transparent signals such as importance, access count, recency, memory type, status, and current layer.

​
Evaluate hierarchy transitions

result = client.evaluate_hierarchy()

print("Promoted:", [memory.id for memory in result.promoted])
print("Demoted:", [memory.id for memory in result.demoted])
print("Archived:", [memory.id for memory in result.archived])
Hierarchy evaluation archives rather than deletes. Every transition adds an audit event.

​
Manual transitions

memory = client.add("Repeatedly useful project fact.", layer="short_term")

episodic = client.promote_memory(memory.id, "episodic")
semantic = client.promote_memory(episodic.id, "semantic")
demoted = client.demote_memory(semantic.id, "episodic")
Invalid layer transitions raise a clear error. Use client.list_by_layer("semantic") to inspect one layer.