Compare
MADB vs a vector database
Vector search finds similar text. Agent memory needs more than similarity.
| Dimension | Vector database | MADB |
|---|---|---|
| Primary job | Find vectors similar to a query | Store and recall agent memory with causes |
| Ranking | Cosine similarity | Similarity + recency + causal proximity + importance + tags |
| Causality | None | caused_by edges + trace_cause |
| "Why" questions | Not supported | Walk the lineage to the root cause |
| Deployment | Often a hosted service | Local-first, on your machine |
A vector database is a great similarity index, and MADB uses semantic similarity too. But similarity alone can’t tell you why an agent acted. MADB adds causal lineage and ranks recall on multiple signals, so the right memory returns and its causes are recoverable.
If you only need nearest-neighbor lookup over embeddings, a vector DB is fine. If you need agent memory — continuity plus accountability — that’s MADB.