A mem0 alternative, with the numbers.
mem0 is a good memory layer and the one most people compare against, so Yohanun was measured against it on the benchmark mem0 reports its own results on: LoCoMo, the long-term conversational memory benchmark from Snap Research. Both systems answered the same 1,540 questions, judged by the same frozen model with identical prompts. Yohanun ran its full production retrieval path, access gate included. mem0 ran its open-source release with its defaults.
Nothing was tuned on either side. That sentence is what the rest of this page exists to earn.
The table
| Category (n) | Yohanun, as shipped | mem0, its defaults | Delta |
|---|---|---|---|
| Overall (1,540) | 62.4% | 48.8% | +13.6 |
| Single-hop (841) | 77.3% | 58.1% | +19.2 |
| Multi-hop (282) | 45.0% | 37.2% | +7.8 |
| Temporal (321) | 46.4% | 39.6% | +6.8 |
| Open-domain (96) | 36.5% | 32.3% | +4.2 |
One thing these numbers cannot do is be compared to anyone's published absolutes, including mem0's own paper, which reports mid-sixties for its system under its own judge. Judges differ; the same mem0 measures 48.8 on this one. Deltas transfer across judges; absolutes do not. Every number here lives on one scale.
The full method, every number, and the two null results that could have been left out are in the benchmark write-up. The entity layer was off in every published number, so the headroom is unread.
What a benchmark cannot measure
Recall accuracy is where the comparison starts, not where the decision is made. mem0 is a memory layer: it stores, extracts and retrieves. Yohanun is governed memory, and three of its properties have no row in any benchmark.
It keeps its history
Nothing is deleted in ordinary operation. A wrong memory is superseded, a finished one closed, both readable by id with their lineage; a question can be asked as of a past date. Docs
It has walls
Compartments, classification and owners on every memory; who is cleared is a platform ledger; the decision is compiled into the retrieval query before the model sees anything. Docs
It leaves a record
One audit row per gated read, written before the answer returns; a sources verdict on every answer; erasure with a plan, a permission and a receipt. Docs
When mem0 is the right choice
If you want an open-source library inside your own process, with no tenancy, no walls and no audit because your application has none to enforce, mem0 is simpler and you should use it. Yohanun is an HTTP service with a key, and it earns that extra hop only when someone will one day ask what the model saw and whether you can prove it.
Try the quickstart: six requests, including the wall and the audit row. Keys are by invite.
Request an invite