Lians vs mem0.
mem0 is the developer-memory API, scored on the same five invariants we score ourselves. Credit where it's strong, receipts where it isn't.
Credit first
Where mem0 is genuinely strong.
mem0 earned its community: excellent developer experience, LLM-driven ADD/UPDATE/DELETE fact management, a hosted platform plus open source, and an open benchmark harness good enough that we score ourselves with it. For personalizing a consumer chat product, it is a fine default.
Recall quality · LOCOMO · run 2026-07-09
On mem0's own harness: Lians 92.9, mem0 91.6.
Regulated-memory eval · last run 2026-07-04
The five invariants.
| Regulated invariant | Lians | mem0 |
|---|---|---|
| Stale revision suppressed | pass | fail |
| Point-in-time (as-of) recall | pass | absent |
| Provable erasure (crypto-shred + cert) | pass | partial |
| Lookahead / backtest guard | pass | absent |
| Audit-state snapshot at T | pass | absent |
| Score (pass = 1, partial = ½) | 5.0 / 5 | 0.5 / 5 |
Where the lanes diverge
Different jobs, different guarantees.
mem0 is ADD-oriented: revised facts accumulate and its LLM decides updates non-deterministically. There is no as-of recall, no event-time model, and deletion removes rows without proving anything. If an examiner asks “what did the agent know on March 1?”, there is no API to answer.
The honest recommendation
Which should you pick?
Choose mem0 for consumer personalization with no audit obligations, big-community support, and the fastest path to “remembers my preferences.” Choose Lians when a regulator, auditor, or backtest can ask what was known when.
A migration guide exists: Lians vs mem0, with migration notes ↗. The write/recall surface maps 1:1; supersession keys replace LLM update heuristics.