Lians vs Letta.
Letta is agent-native memory (MemGPT lineage), scored on the same five invariants we score ourselves. Credit where it's strong, receipts where it isn't.
Credit first
Where Letta is genuinely strong.
Letta pioneered self-editing memory inside a complete agent framework. Agents curate context blocks and search an archival store.
Regulated-memory eval · last run 2026-07-04
The five invariants.
| Regulated invariant | Lians | Letta |
|---|---|---|
| Stale revision suppressed | pass | partial |
| 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 | 1.0 / 5 |
Where the lanes diverge
Different jobs, different guarantees.
Letta overwrites memory without validity windows or time-aware archival search. Lians closes revisions and reconstructs past state with recall_at().
The honest recommendation
Which should you pick?
Choose Letta for a self-managing agent runtime. Choose Lians for auditable memory beneath any framework.