memory-systems

Start with file-system memory storing facts as timestamped JSON, then escalate to a vector store like Mem0 when you need semantic search. Add a temporal knowledge graph such as Zep/Graphiti only when you need relationship traversal or time-travel queries.

Choose Mem0 for fast time-to-production with managed infrastructure, Zep/Graphiti for bi-temporal modeling and relationship reasoning, and Cognee for dense multi-layer semantic graphs with customizable ingestion pipelines. Letta fits agents needing deep self-introspection over their own memory.

A temporal knowledge graph stores facts with valid_from and valid_until timestamps so agents can answer point-in-time queries like where a user lived on a specific date. Use one when facts change over time and stale data would poison agent behavior.

Benchmark against LoCoMo for long-conversation retention, LongMemEval for long-term memory, and DMR for retrieval accuracy. Treat results as dated evidence and re-measure after any architecture change rather than trusting absolute rankings.

Stale retrieval happens when facts lack temporal validity tracking or consolidation never runs. Track valid_from and valid_until on mutable facts, prefer the most recent fact on conflict, and consolidate periodically by invalidating rather than deleting old entries.

Skip vector databases during prototyping since plain files with naming conventions can match specialized tooling on some benchmarks. Semantic search also degrades on multi-hop reasoning queries, where graph traversal is the better fit.