New In-Memory Database Purpose-Built for AI Agents

Why it matters

Neo4j has released Relate, an open source, in-memory, in-process embedded database written in Rust and aimed specifically at AI agents and agentic systems. The launch reflects a shift in database demand: Gartner forecasts that machine users, not humans or basic assistants, will dominate DBMS workloads within a few years, generating far more interactions than people do and driving infrastructure and spending growth in the process. Relate is designed around autonomous execution loops rather than traditional human-driven queries. It auto-detects agent harnesses, returns structured errors with explicit retryable flags so agents can self-correct, and caps result rows with truncation markers to keep responses within context limits. It connects to agents through a Model Context Protocol server, supports Cypher queries with vector similarity and graph traversal in a single statement, and runs in notebooks, browsers, containers, and on edge devices, with optional durable disk storage. This signals graph vendors competing on agent-native ergonomics.