Head to head

flmnt vs Letta

Letta is agent framework, and it does recall well — remembering what was said is a solved problem. The lines below are about currency: whether the decision you get back is the one that holds now. Where a claim has a number, it traces to our public benchmark.

Where memory comes from
How memories are created
flmnt:
Authored by your agents as they work
Letta:
Agent-managed notes
Decisions, causes & mistakes as first-class types
flmnt:
Letta:
Nothing guessed after the fact
flmnt:
Letta:
partial
agent-written
Currency — what's true now
Serves the current decision after a reversal
flmnt:
Letta:
Supersession test (no textual tell)
flmnt:
1.00
Letta:
Reversed decisions served as history, never truth
flmnt:
Letta:
Causal lineage — the why behind each decision
flmnt:
Letta:
Tracks whether mistakes stop repeating
flmnt:
Letta:
Recall — what was said
Semantic recall over history
flmnt:
Letta:
Session-to-session persistence
flmnt:
Letta:
Economics
Cost per query as the corpus grows
flmnt:
Flat — $0.0077 measured
Letta:
Grows with context
Live ROI ledger, per deployment
flmnt:
Letta:
Trust & fit
Public benchmark, run every release
flmnt:
Letta:
MCP-native — no SDK, no rewrite
flmnt:
Letta:
Framework adoption
Immutable, replayable record
flmnt:
Letta:

Entries reflect Letta's published approach as of July 2026 — corrections welcome, and the full matrix covers every tool side by side.

See the full comparison across every tool →