Use this table when deciding where data should live:
DataMongoDBQdrantmem0
User profiles, org settings
Messages, tasks, channels
Embeddings / semantic retrieval chunksoptional metadata✓ vectors
Long-horizon conversation summaries (feature-dependent)backup refs✓ when enabled
Audit / compliance records

How to read each store

  • Authoritative state — Read from MongoDB for anything that must be transactionally correct.
  • Similarity search — Use Qdrant with organization filters.
  • Long conversational memory — Use mem0 when enabled; otherwise rely on recent Mongo history or product-specific fallbacks.

Retention and lifecycle

Decide per deployment how long data is kept: Mongo backups, Qdrant collection policies, mem0 vendor rules. Record operator expectations in Operations runbooks.

Performance

Vector upserts and queries add latency; mem0 adds network hops. Load-test chat and admin flows before large pilots.