DEV Community

News
News

Posted on

Neo4j launches Relate, an open-source Rust graph database built directly into AI agents

Infographic cover

๐Ÿ“… November 11, 2026 โ€ข Source: Neo4j blog

At the NODES 2026 conference Neo4j launched Relate, an AI-native, open-source graph database built in Rust for AI agents and agentic systems. It runs in memory, inside the process of the host application, and ships under the Apache 2.0 license โ€” install from relatedb.ai.

Startup takes milliseconds and in-memory reads come in under a millisecond. The engine has an optional fsync write-ahead log for durable stores, and storage can be backed by Parquet, JSON or Markdown files โ€” a single file that is easy to share or check into GitHub. Compliance is a headline claim: 100 percent openCypher and more than 97 percent GQL, asserted by more than 30k test scenarios across openCypher, Cypher 25 and GQL.

Relate is agent-native by design. It auto-detects agent harnesses, agent mode returns compact TOON output with capped rows, structured errors carry retryable flags so agents can self-correct, and a built-in relate agent-context documents the installed version. Connectivity includes a built-in MCP server, a dedicated CLI, and Bolt plus HTTP Query API v2 endpoints for existing graph drivers and frameworks. There are libraries for Python, JavaScript, Rust, Java, Go and Swift, and a WebAssembly build that runs in the browser.

One Cypher statement can combine vector similarity search and graph traversal โ€” the pattern behind GraphRAG and multi-hop reasoning. When a project outgrows local runs, the same Cypher works in Neo4j Aura: pull a subgraph locally, push results back. Neo4j highlights typical uses: agent coordination with a shared record of work, per-agent memory, sandboxes (one file, offline) and codebase context graphs.

Why it matters: agents need fast, local, structured state, and an embedded graph database that speaks Cypher, exposes MCP and fits in one file targets exactly that. Neo4j cites Gartner: machine users are projected to make up 75 percent of database workloads by 2029, up from 10 percent in 2025. If that curve holds, databases designed for machine-first access stop being a niche โ€” and Relate is Neo4j's bid to own that layer from the laptop up.

Sources: neo4j.com - businesswire.com - dbta.com

Key figures

IMPACT checklist


Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky

AI Automation

Top comments (0)