Remnic vs Zep.
Zep is the closest system to Remnic on temporal reasoning, and the furthest on operational weight. The honest comparison is about what you are willing to run.
What Zep is
Zep, with its open-source Graphiti engine, is the reference architecture for temporal agent memory. Every fact is a graph node with a validity window: when it became true and when it was superseded. That bi-temporal posture makes "when did this change and why" queries first-class. The cost is operational: Graphiti wants a Neo4j, FalkorDB, or Kuzu instance, and managed features live behind the Zep cloud subscription.
Where Zep wins
- Deep temporal graph queries over long fact histories, natively.
- A managed cloud path if you want someone else running the infrastructure.
- A mature graph data model for teams already invested in graph tooling.
Where Remnic differs
- Temporal supersession without a database. New facts with a matching supersession key mark the old fact superseded at write time. An overnight contradiction scan pairs semantically similar actives, classifies them with an LLM judge, and queues conflicts for your resolution. No auto-delete.
- Files over nodes. Memories are markdown with YAML frontmatter. The search index is downstream and rebuildable. Nothing about your memory requires operating Neo4j.
- Multi-host by design. One store shared by Claude Code, Codex CLI, Cursor, Pi, Hermes, OpenClaw, and any MCP client.
- A published correction protocol. MemCorrect stresses whether corrections stick, resurrect, or take collateral damage. See benchmarks.
Switching
There is no automated Zep importer today. Zep memories live in your graph instance, and export shape varies by deployment. If you want one, the issue tracker is the right place to ask: github.com/joshuaswarren/remnic/issues. Remnic starts building fresh memory from your first session either way.