ArangoDB vs Neo4j
ArangoDB vs Neo4j: side-by-side comparison of two knowledge-graph platforms systems — architecture, taxonomy, license, pricing, MCP/A2A support, and direct edges.
Cost & capability
| ArangoDB | Neo4j | |
|---|---|---|
| Cost tier | mid | free |
| $/Mtok input | — | 0 |
| $/Mtok output | — | 0 |
Where they differ (13)
Rows where both sides have data and the values disagree — the shortlist of dimensions that actually distinguish these two systems.
| ArangoDB | Neo4j | |
|---|---|---|
| Cost tier | mid | free |
| Type | Multi-model: graph + doc + KV + vector | Property graph + vectors + Aura Agent |
| Created | 2014 | 2024-12 |
| License | BSL-1.1 (source-available; each release converts to Apache-2.0 four years after release) | GPL-3.0 (Community Edition); Enterprise Edition proprietary |
| GitHub | arangodb/arangodb — ~14.3k stars, active (commit date not captured) | neo4j/neo4j — ~17.2k stars, active (default branch 2026.07) |
| Pricing | Pay-per-use | Free + paid |
| Funding | Series B · 2021-10 | $325M total $2.0B val Series F · 2021-06 |
| Backend storage | custom (RocksDB-backed multi-model) | custom (native graph store) |
| Deployment | Both | Both (self-hosted + managed cloud) — Community/Enterprise self-managed; Neo4j Aura cloud |
| API surface | REST (HTTP), SDK: many | Bolt, REST, GraphQL, Cypher, SDK: many |
| Multi-tenancy | hard-isolation | AuraDB Virtual Dedicated Cloud / AuraDS Enterprise: dedicated AWS/Azure/GCP account/subscription/project per customer; PrivateLink supported |
| MCP | yes — community ravenwits/mcp-server-arangodb | yes — official neo4j/mcp (product team); Neo4j Labs also ships neo4j-contrib/mcp-neo4j |
| OpenTelemetry | first-class | first-class — Neo4j metrics + OTel |
At a glance
| ArangoDB | Neo4j | |
|---|---|---|
| Section | Knowledge-graph platforms | Knowledge-graph platforms |
| Tier | T1 | T1 |
| Type | Multi-model: graph + doc + KV + vector | Property graph + vectors + Aura Agent |
| Created | 2014 | 2024-12 |
| Latest release | — | 2026.07 CalVer series (current default branch; exact patch and date not captured) |
| License | BSL-1.1 (source-available; each release converts to Apache-2.0 four years after release) | GPL-3.0 (Community Edition); Enterprise Edition proprietary |
| GitHub | arangodb/arangodb — ~14.3k stars, active (commit date not captured) | neo4j/neo4j — ~17.2k stars, active (default branch 2026.07) |
| Pricing | Pay-per-use | Free + paid |
| Funding | Series B · 2021-10 | $325M total $2.0B val Series F · 2021-06 |
| Backend storage | custom (RocksDB-backed multi-model) | custom (native graph store) |
| Deployment | Both | Both (self-hosted + managed cloud) — Community/Enterprise self-managed; Neo4j Aura cloud |
| API surface | REST (HTTP), SDK: many | Bolt, REST, GraphQL, Cypher, SDK: many |
| Embedding | BYO | BYO |
| Multi-tenancy | hard-isolation | AuraDB Virtual Dedicated Cloud / AuraDS Enterprise: dedicated AWS/Azure/GCP account/subscription/project per customer; PrivateLink supported |
| MCP | yes — community ravenwits/mcp-server-arangodb | yes — official neo4j/mcp (product team); Neo4j Labs also ships neo4j-contrib/mcp-neo4j |
| A2A | no Google A2A (Agent2Agent) integration documented as of 2026-05. | no Google A2A (Agent2Agent) integration documented as of 2026-05. |
| OpenTelemetry | first-class | first-class — Neo4j metrics + OTel |
| Optimised for | relationship modeling + reasoning + governance over pure vector | relationship modeling + reasoning + governance over pure vector |
| Anti-fit | not for purely-vector or simple-RAG use cases (graph adds setup cost) | not for purely-vector or simple-RAG use cases (graph adds setup cost) |
Pros & cons
ArangoDB
Pros: Multi-model (graph + document + KV) in one engine — useful when memory crosses paradigms without managing multiple databases.
Cons: Multi-model adds complexity vs single-purpose engines; smaller community than Neo4j or MongoDB.
Neo4j
Pros: Largest graph database community by orders of magnitude; Cypher is the de-facto graph query language; strong native AI/RAG integrations.
Cons: Property-graph model can be awkward for RDF/semantic-web workloads; clustering and scale-out can get expensive.