Memgraph

Memgraph is an in-memory, ACID-compliant graph database written in C++ that speaks openCypher over the Bolt protocol, so most Neo4j client code ports across unchanged. It targets teams building GraphRAG pipelines, agent memory, fraud detection and network analysis, where sub-millisecond multi-hop traversal matters more than storing a graph larger than available RAM.

Memgraph is an in-memory, ACID-compliant graph database written in C++ that now positions itself as "the graph engine for AI context." It speaks openCypher over the Bolt protocol, making it a near drop-in swap for existing Neo4j clients, but it keeps the working graph in RAM rather than paging it from disk. That single architectural choice is behind both its headline claims — sub-millisecond multi-hop traversals, roughly 1,000 read and write transactions per second, graph sizes the vendor scopes from 100 GB to 4 TB — and its main constraint, since the dataset has to fit in memory and the documented guidance is to provision about twice as much RAM as data. Durability comes from write-ahead logging plus periodic snapshots, with replication and automatic failover for high availability, and an on-disk transactional storage mode exists for datasets that exceed RAM at explicitly slower performance. The MAGE library ships 40-plus graph algorithms including PageRank, community detection and centrality; native stream connectors read from Kafka, Pulsar and Redpanda, alongside Parquet and JSONL loading. The AI push is the recent story: built-in vector and text indexes let one query combine similarity search with full graph traversal, which is the substance of Memgraph's GraphRAG pitch, and a documented AI-memory pattern stores semantic, episodic and procedural agent memory in a single graph with inspectable execution traces. Memgraph Lab handles visual query and schema exploration. Named users include NASA, which publicly moved off Neo4j citing cost, plus Cedars-Sinai, Capitec Bank, Microchip, Sayari and Skillsoft.

Category

Data & Analytics

Pricing

Freemium, Subscription, Usage-based, Contact for pricing

Target Market

CTOs, Data Engineers, Enterprise Developers, ML/AI Engineers, Platform Engineers

Deployment

Self-hosted, Hybrid, Cloud-first

Founded

2016

Headquarters

London, United Kingdom

• ✓
  In-memory C++ storage engine

  Keeps the working graph in RAM with ACID transactions, delivering sub-millisecond multi-hop traversals instead of disk-paged reads.

• ✓
  openCypher and Bolt compatibility

  Speaks the same query language and wire protocol as Neo4j, so existing drivers, tooling and queries largely port across without a rewrite.

• ✓
  Native vector and text indexes

  Combines similarity search with graph traversal inside a single query, removing the round-trip between a separate vector store and the graph.

• ✓
  MAGE algorithm library

  Ships 40-plus graph algorithms including PageRank, centrality and community detection, callable directly from Cypher without exporting data.

• ✓
  Streaming and bulk ingestion connectors

  Reads directly from Kafka, Pulsar and Redpanda for live graphs, plus native Parquet and JSONL loading for bulk backfills.

• ✓
  Enterprise access control and multi-tenancy

  Fine-grained RBAC, label-based access, multiple roles per user, SSO via Entra ID, Okta, OIDC and SAML, LDAP/PAM and query audit logging.

• ✓
  Memgraph Lab

  Visual query editor, schema explorer and result graph visualisation, which shortens the debugging loop on unfamiliar or evolving graph models.

• •
  GraphRAG retrieval for enterprise assistants

  Traverse a knowledge graph across multiple hops to assemble context that pure vector similarity misses, then return it inside a single low-latency query.

• •
  Real-time credit and risk scoring

  Capitec Bank runs scoring across more than 3.5 million clients daily, completing the pass in roughly two hours on Memgraph.

• •
  Agent memory and reasoning state

  Store semantic, episodic and procedural agent memory in one graph with inspectable execution traces, so agent decisions can be audited after the fact.

• •
  Fraud ring and anomaly detection

  Detect connected-entity patterns inside a live transaction path where a disk-backed graph traversal would blow the latency budget.

• •
  Network and dependency analysis

  Model infrastructure, supply chain or identity dependency graphs and run centrality and community algorithms directly against the live dataset.

Community Edition is free but ships under BSL 1.1, which is source-available rather than OSI-approved open source and explicitly bars offering Memgraph as a service. Enterprise is quoted against memory capacity rather than seats or cores, and an independent 2026 comparison reports list starting around $25,000/year for 16 GB. Two Enterprise variants exist: the AI Platform tier meters graph data only and leaves vector indexes unlicensed, while Standard meters graph data and vector indexes combined. Memgraph Cloud bills pay-as-you-go on 1-32 GB instances after a two-week trial. Budget the RAM alongside the licence — it is usually the larger line.

This page was written from 8 sources, 5 on domains other than memgraph.com.

1. 1.memgraph.com — pricingvendor
2. 2.memgraph.com — install memgraphvendor
3. 3.memgraph.com — authentication and authorizationvendor
4. 4.github.com — memgraph
5. 5.puppygraph.com — memgraph vs neo4j
6. 6.arcadedb.com — neo4j alternatives in 2026 a fair look at the open source op
7. 7.tracxn.com — funding and investors
8. 8.hn.algolia.com — hn.algolia.com