Unison vs Zep - Team Brain vs Temporal Knowledge Graph

Zep builds a temporal knowledge graph per user; Unison builds one per project, shared by every agent and human on the team. Where each fits.

Zep and Unison agree on more than most pairs in this category: both reject vector-soup memory, both model knowledge as a temporal graph where facts change over time. The difference is what the graph is about.

Zep's graph is about a user. Sessions, messages, and entities accumulate per user; the canonical question is "what does this user's history imply right now?"

Unison's graph is about a project. Many agents and humans write into one workspace; teams and private lanes coexist; the canonical question is "what does the team know - and which of it does this session need?"

	Zep	Unison
Graph scope	per user	per workspace, with /workspace/teams/ and /private/ lanes
Temporal model	fact validity intervals	bitemporal facts + supersession + `asOf` reads
Multi-agent writes	n/a (single-user design center)	conflict review queue, actor delegation
Recall surface	search + memory APIs	one-call `contextMd` with abstention flag
Document tier	n/a	scope-by-path markdown filesystem (+ unison-fs mount)
Agent messaging	n/a	backchannel

If your product is "an assistant that remembers its user," Zep is a reasonable choice. If your problem is "five sessions across three harnesses and two people keep re-learning and contradicting each other," that's the problem Unison is shaped for.