Supermemory is the best-performing competitor in PrecisionMemBench. Memory graph with dynamic dreaming. Mean precision: 0.43. Still returns 2-3 irrelevant beliefs per correct one.
Supermemory is architecturally more sophisticated than Mem0, Zep, or Hindsight. Its memory graph and dynamic dreaming mechanism provide better signal than raw cosine similarity. 17 active retrieval passes is meaningful; no other competitor scores above zero.
But 0.43 precision still means that for every correct belief returned, there are 1-2 irrelevant ones alongside it. On 26 of 43 precision-assertion cases, the system fails. The memory graph helps, but it does not resolve the fundamental problem: semantic proximity within a domain-specific corpus cannot be discriminated by any similarity-based mechanism.
0.43 precision means your model still has to sort signal from noise on most queries. Every irrelevant belief in context is tokens you are paying for and latency you are waiting on. Tenure's structural approach (alias-weighted BM25 + hard scope isolation) does not degrade with corpus growth. It gets more precise as the alias set expands.
Supermemory's p50 retrieval latency of 819.48ms is 84x slower than Tenure's 9.77ms. Session-turn latency averages 867.83ms. This is the cost of running a memory graph traversal and dynamic dreaming pipeline on every query.
Supermemory is a cloud-first SaaS product. Your data leaves your machine. Tenure runs entirely on localhost. Nothing ever leaves your machine. No account required.
| Property | Tenure | Supermemory |
|---|---|---|
| Mean retrieval precision | 1.00 | 0.43 |
| Active retrieval passes | 43/43 | 17/43 |
| Total passes (77 non-session) | 77/77 | 44/77 |
| Session turns passed | 12/12 | 2/12 |
| Retrieval latency (p50) | 9.77ms | 819.48ms |
| Session latency (p50) | 47.79ms | 867.83ms |
| Drift score | 0.00 | 0.17 |
| Runs locally | Always | Cloud SaaS |
| Account required | No | Yes |
| Data leaves machine | Never | Always (cloud) |
| Scope isolation | Hard filter | Partial (graph-based) |
| Per-turn injection audit | Yes | No |
| Pricing | Free (MIT) | $19-$399+/mo |
| License | MIT | Proprietary |
Full methodology: arXiv:2605.11325. Dataset: HuggingFace. Live leaderboard: HuggingFace Spaces.
No cloud. No account. No monthly bill. Thirty seconds to install. Runs on your machine forever.