AppRisingofficial publisher

Zep / Graphiti

App : A product you use directly

Temporal knowledge graph memory for agents, with Graphiti in open source.

The choice when memory must understand time and relationships, not just retrieve facts.

Official site GitHub · 32,187 ★

Overall score

7.4 /10

Security: one caveat, see the card

Utility

8.0

Reliability

10.0

Security

One caveat

Ease

Technical · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

AI agents forget what they have learned as soon as a conversation ends, and traditional memory systems cannot handle information that changes over time. Zep and Graphiti address this by building a temporal knowledge graph: every fact is timestamped, and when information changes, the old version is invalidated without being deleted.

In practice, Graphiti ingests text or structured data (JSON, conversation histories, business data) and transforms it into entities, relationships, and facts linked together. Each fact retains a validity window and full traceability back to its raw source. Search combines semantic embeddings, keywords (BM25), and graph traversal for precise queries in under one second.

Graphiti is the open-source layer, self-hosted with Neo4j, FalkorDB, or Amazon Neptune. Zep is the managed version, aimed at teams that want production-ready infrastructure: sub-200ms latency, attribute-based access control, retention policies, audit logs, and deployment inside your own cloud if needed.

This tool is intended for developers and technical teams building AI agents that require persistent, personalized, and time-consistent memory. It is not suited for non-technical users or projects that do not need to track how information evolves over time.

What it offers

·Temporal knowledge graph: every fact has a validity window and is automatically invalidated when newer information contradicts it.
·Full provenance: every entity and relationship traces back to the source episode (raw data) that produced it.
·Incremental ingestion: new data integrates immediately without full graph recomputation.
·Hybrid retrieval: combines semantic embeddings, keyword search (BM25), and graph traversal for precise results.
·Prescribed or learned ontology: entity and edge types can be defined via Pydantic models or emerge automatically from data.
·Multi-graph-database compatibility: Neo4j, FalkorDB, and Amazon Neptune are supported for self-hosting.
·Attribute-based access control, retention policies, and audit logs (paid): data governance at the data layer in Zep.
·Bring Your Own Cloud (BYOC) deployment with sub-200ms retrieval latency regardless of graph size (paid).

Our tips

By default, concurrency is capped at 10 simultaneous operations (SEMAPHORE_LIMIT variable) to avoid LLM provider 429 errors; increase this value if your provider allows it to speed up ingestion.
Graphiti works best with LLMs that support structured output (OpenAI, Anthropic, Gemini); using smaller models or other providers may cause ingestion failures.
To get started quickly without configuring Neo4j, launch FalkorDB via Docker using the provided command and use the quickstart example included in the repository.
Define your entity and edge types via Pydantic models from the start if your domain is well known; let the ontology emerge from data if you are exploring a new domain.
Avoid Kuzu as a graph database backend for new projects: this backend is deprecated and will be removed in a future release.

Watch out

Self-hosting Graphiti requires managing the graph database, developer tooling, and retrieval performance independently, which demands significant technical investment.
LLM providers without structured output support may produce incorrect schemas and ingestion failures, limiting the choice of usable models.

In practice

Price: Freemium
Commercial use: Commercial use: yes
Available on: web, docker, cli
Licence: Apache-2.0

Security and data

Rating: Security: one caveat, see the card
Your data: Local or cloud depending on use
Maintenance: Active
Last release: 20 October 2026

Links and repository checked on 25 October 2026 · added on 1 August 2024

How to install Zep / Graphiti

1. Verify that Python 3.10 or higher is installed on your machine. 2. Install Graphiti via pip: pip install graphiti-core (or with FalkorDB: pip install graphiti-core[falkordb]). 3. Start a compatible graph database: for example FalkorDB via Docker with docker run -p 6379:6379 -p 3000:3000 -it --rm falkordb/falkordb:latest, or Neo4j Desktop. 4. Set your OpenAI API key in the OPENAI_API_KEY environment variable (or the key for your chosen LLM provider). 5. Follow the quickstart example provided in the examples/quickstart folder of the GitHub repository to initialize indexes and test ingestion. 6. Never include your API key in source code; use environment variables or a secrets manager.

pip install graphiti-core

In the same category

Whole category Feed (memory, context) →