Neo4j launches Relate embedded database for AI agents

The open source database supports AI agents with embedded graph capabilities, built-in connectivity and optional disk persistence, allowing developers to run agent workloads across notebooks, browsers, containers, mobile devices and edge environments using existing tools.

Neo4j has released Relate, an open source embedded database designed for AI agents and agentic systems. Built in Rust, the database runs in memory and within an application’s own process, with optional disk storage for persistence.

The company said Relate enables AI coding agents to create a database and begin building applications in milliseconds. Released under the Apache 2.0 licence, it is designed for agent sandboxes, agent memory and deployments across mobile and edge devices.

Graph capabilities for agent workflows

Relate supports multihop reasoning and graph retrieval-augmented generation, or GraphRAG, which Neo4j said can help AI agents produce more accurate results and reduce hallucinations.

“Relate is the ultimate agent-native database,” said Sudhir Hasbe, president of technology and CPO at Neo4j. “Built in Rust for maximum speed and efficiency, it acts as a lightweight, in-memory store with highly durable disk storage, when needed, that can run anywhere. From seamless agent connectivity via MCP to multi-hop reasoning through full Cypher support, Relate provides a unique out-of-the-box solution designed to run in-process for agent sandboxes and agent memory as well as on mobile or edge devices.”

The database automatically detects agent harnesses and returns structured errors with explicit flags indicating whether an operation can be retried. These are intended to help agents handle exceptions and self-correct without custom parsing logic.

Relate also limits the number of rows returned and includes a marker when results are truncated, helping keep retrieval responses within an agent’s context window.

Connectivity across development environments

Relate can run across Jupyter notebooks, web browsers, software containers and edge devices. It includes a Model Context Protocol (MCP) server and a dedicated command-line interface for agent connectivity and direct execution.

The database also exposes Bolt and HTTP Query API v2 endpoints, supporting integration with existing graph drivers, orchestration frameworks and developer tools.

In-memory execution and optional persistence

According to Neo4j, Relate delivers sub-millisecond in-memory reads to support frequent read and write operations in agent workloads. A single Cypher statement can combine vector similarity searches with graph traversals.

The database supports transactional integrity through its in-memory engine, with an optional write-ahead log synchronised to disk for durable storage. Developers can also flush database state to disk for longer-term persistence.

Neo4j said Relate is fully compliant with openCypher and more than 97% compliant with Graph Query Language (GQL). The database is available through relatedb.ai.