Memtro vs Graphiti
An open-source temporal knowledge-graph library, versus a hosted product built on a graph in Postgres.
This comparison describes Graphiti as we understand it at the time of writing from its public repository. Projects change; check the current details before deciding.
Graphiti is an open-source library for building temporal knowledge graphs for agents: you feed it episodes (conversations, documents, structured data) and it extracts entities and relationships with validity intervals into a graph database, then serves hybrid retrieval.
Memtro uses similar ideas (typed, time-aware facts, entities with aliases, supersession, hybrid retrieval) but as a finished product: an endpoint, a dashboard, connectors, routing, agent mode and jobs, on a graph stored in Postgres.
Where they differ
| Memtro | Graphiti | |
|---|---|---|
| What it is | A hosted product | A library you build with |
| Graph store | Postgres with pgvector, recursive CTEs for traversal | A graph database you run (for example Neo4j or FalkorDB) |
| Extraction | Built in, gated by signals, with provenance and per-fact scope | Built in, configured by you |
| Data in | Chat, fifteen read-only connectors, hourly syncs, imports | Episodes your code sends |
| Interfaces | OpenAI-compatible API, MCP server, dashboard | Python API |
| Agent runtime | Agent mode, engines, jobs | Your own |
When to choose which
Choose Graphiti when you want to design the memory of an agent you are building and are happy to run a graph database. Choose Memtro when you want the memory, connectors and agent runtime ready to use, and are happy with Postgres as the single store. If graph traversal ever outgrows SQL, Memtro's store is isolated so a graph database can be swapped in.