---
title: "past.dev vs Graphiti"
description: "Compare Graphiti and past.dev by licence, what each one asks you to operate, ingestion format, deployment, and recall output."
canonical: https://past.dev/vs/graphiti
last-updated: 2026-10-26
---
# past.dev vs Graphiti

Source: https://past.dev/vs/graphiti

Graphiti is the open-source framework behind Zep, under Apache-2.0. You install it as a library and run it on a graph database you operate yourself. past.dev provides a managed or self-hosted JSON API for timestamped source ingestion and grouped, ranked recall with dated source excerpts. Choose between an engine you operate and an API contract your application calls.

## What Graphiti is

[Graphiti](https://github.com/getzep/graphiti) is an Apache-2.0 framework for temporal knowledge graphs, maintained by the Zep team. It ingests episodes, the raw stream as it arrived, and derives entity nodes and relationship edges from them. Each fact carries a validity window: when it became true, and when it was superseded. A superseded fact is invalidated rather than deleted, so the history stays answerable.

It installs as `graphiti-core` and runs on a graph database you provide: Neo4j, FalkorDB or Amazon Neptune. It calls a language model to extract entities and edges, so a provider key is required; OpenAI is the default, with Anthropic, Gemini, Groq and OpenAI-compatible endpoints also supported. The repository also ships an MCP server and a FastAPI REST service over the graph.

## Graphiti and Zep are not the same thing

Graphiti is the engine, and it is the part you can read, fork and run. Zep is the managed service from the same team, built on it. If you are comparing one managed service against another, the page you want is [past.dev vs Zep](/vs/zep). If you are deciding whether to operate the engine yourself, stay here.

## What past.dev is

past.dev accepts raw text and its original timestamp without a required schema, and the timestamp can be backdated so an import does not land as today. The same public routes are available through the managed service and the self-host bundle.

`/recall` returns ranked documents with occurrence dates and source excerpts under a token budget. The application decides whether those documents establish an answer.

## Architecture differences

|  | Graphiti | past.dev |
| --- | --- | --- |
| Licence | Apache-2.0 | Commercial, with a self-host image |
| What you operate | The library, plus a graph database | Nothing on the managed service; the published deployment bundle when self-hosted |
| Storage underneath | Neo4j, FalkorDB or Amazon Neptune | Not part of the public API contract |
| What you send | Episodes: messages, text or JSON | Text plus the timestamp it happened, backdatable |
| Model provider key | Required, for entity and edge extraction | Not part of the API contract on the managed service |
| Read surface | Graph search, an MCP server and a REST service | `/recall`: ranked documents with dated source excerpts |
| Abstention | Your code decides | Your code evaluates returned documents |
| Cost visibility | Your infrastructure bill, plus model usage | Accounted per call on the console dashboard |

## How to choose

Choose based on the requirements below.

- **Choose Graphiti** if you want the engine under an open-source licence, if you already run Neo4j, FalkorDB or Neptune, or if you want to write the traversals yourself.
- **Choose past.dev** if you want a memory API with timestamped ingest, identities, audiences and source-backed recall, and nothing new to operate.
- **Both** record time on facts rather than overwriting them. [Knowledge graphs for LLM applications](/knowledge-graph-for-llm) explains what that buys you.

## What this page does not claim

This page contains no head-to-head scores. Our evaluation uses BEAM and LoCoMo with community-corrected answer keys, with full-context and plain-RAG baselines and several judges. The method is published on [benchmarks](/benchmarks). We will publish product comparisons after documenting a reproducible configuration for the other product.

## Related

- [past.dev vs Zep](https://past.dev/vs/zep)
- [Knowledge graphs for LLM applications](https://past.dev/knowledge-graph-for-llm)
- [What is agent memory?](https://past.dev/what-is-agent-memory)
- [How the pipeline works](https://past.dev/docs/memory-api/how-it-works)
- [How we measure memory](https://past.dev/benchmarks)