Head to head

flmnt vs Zep

Zep is temporal graph, 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
Zep:
Extracted into a graph
Decisions, causes & mistakes as first-class types
flmnt:
Zep:
partial
entities & relations
Nothing guessed after the fact
flmnt:
Zep:
Currency — what's true now
Serves the current decision after a reversal
flmnt:
Zep:
partial
time-stamped facts
Supersession test (no textual tell)
flmnt:
1.00
Zep:
Reversed decisions served as history, never truth
flmnt:
Zep:
Causal lineage — the why behind each decision
flmnt:
Zep:
partial
graph edges
Tracks whether mistakes stop repeating
flmnt:
Zep:
Recall — what was said
Semantic recall over history
flmnt:
Zep:
Session-to-session persistence
flmnt:
Zep:
Economics
Cost per query as the corpus grows
flmnt:
Flat — $0.0077 measured
Zep:
Grows with graph
Live ROI ledger, per deployment
flmnt:
Zep:
Trust & fit
Public benchmark, run every release
flmnt:
Zep:
partial
published evals
MCP-native — no SDK, no rewrite
flmnt:
Zep:
SDK / API
Immutable, replayable record
flmnt:
Zep:
partial

Entries reflect Zep'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 →