Comparison
Contextely vs Zep
Zep models how a fact changed over time. Contextely does not try to: it holds one condensed current answer and re-fetches it from the source the moment its TTL expires.
What Zep is
A memory service built on a temporal knowledge graph, tracking how facts about users and sessions change over time so an agent can reason about what was true when.
Choose Zep when
- You need to reason about the HISTORY of a fact: what it was in March versus now.
- Your data is naturally a graph of entities and relationships you want to traverse.
- Chat-session memory with temporal validity is the core problem you have.
Choose Contextely when
- You do not need history. You need the CURRENT value, and you need it to be right.
- The authoritative copy already lives in a database you own, and you would rather re-read it than model its history.
- Access control per asker is a hard requirement rather than an application concern.
| Dimension | Zep | Contextely |
|---|---|---|
| Model | Temporal knowledge graph with bi-temporal edges | Condensed objects with a freshness contract each, linked by shared entities and topics. A graph you can walk, not one that models how a fact changed over time |
| Source of truth | The graph, built from ingested episodes | Your systems of record. The store is explicitly a cache with rules |
| Staleness handling | Memory is written when you write it; keeping it current is your pipeline's job | Every object carries a TTL and is re-fetched from its system of record at query time when it expires |
| Per-asker permissions | Not a built-in concept. The application above it is expected to scope what it stores and retrieves | Entitlement scopes applied inside the retrieval scoring function, before synthesis; unentitled objects score zero |
| Self-host | Open source, self-hostable | Open self-host path, Docker, no feature gating |
Memory objects
5 of 100Acme Industrial: renewal terms
fetched 4s ago
Q3 churn commitments to Acme
fetched 3m ago
Support SLA by plan tier
fetched 2h ago
Northwind pricing sheet v4
fetched 1h 12m ago
Engineering comp bands 2026
fetched 6h ago
One object is past its TTL. The next query that selects it re-reads the record from its source before it answers, rather than serving what is stored.
What the store looks like
A cache with rules, not a second source of truth
Every object carries the TTL its source set and the moment it was last read. Anything past its window is re-read from the system of record before it is allowed to answer, so the store never quietly becomes the thing your company believes.
Free for 500 retrievals a month, and self-hostable with no limits.