Skip to content
Rush Commerce
Tools & Teardowns3 min read

Neo4j Relate: an embedded graph database for AI agents

Neo4j Relate is an open-source embedded graph database for AI agents, written in Rust with MCP included. Test it for agent memory, and read the install script first.

Neo4j Relate is an embedded graph database for AI agents, and it runs with no server and no account. Neo4j launched it on November 11 at its NODES 2026 conference. It is open source, it is a new Rust implementation, and it runs as a library inside your application. The pitch is agent memory with structure: the agent stores what it learns as nodes and relationships, then queries them with Cypher. The install path is one curl | bash line that the project page tells your agent to run. Read that script before the agent does.

What actually happened

The details come from Neo4j's announcement and the project site:

  • The form. A library in your app, a CLI in a sandbox, or a WebAssembly build in the browser. The site says there is no signup and no server to deploy.
  • The storage. In memory for a session you throw away. Or a directory or a single file, where each write is saved to disk as a transaction. Neo4j says storage can also be backed by Parquet, JSON or Markdown files.
  • The query language. Cypher. Neo4j says Relate runs more than 30,000 test scenarios for openCypher, Cypher 25 and GQL. One statement can follow relationships and run vector and full-text search together.
  • The agent parts. relate serve exposes a database as MCP tools to read the schema and run queries. relate skill install adds a skill for coding agents.
  • The bindings. Python, JavaScript, Rust, Java, Go and Swift.
  • The license. Apache-2.0, per the project site.
  • The speed claim. "Millisecond startup and submillisecond queries" for short sessions. That is Neo4j's claim. The post gives no benchmark or method.
  • The upgrade path. If the project grows past local use, Neo4j points you to Neo4j Aura, its hosted product. The same Cypher runs there.

Why Neo4j Relate matters for your business

Most agent memory today is a vector store or a folder of Markdown notes. Both are weak at questions with more than one step, such as "which customers bought the part that this supplier recalled?" A graph answers that directly. We saw the same result when a code graph cut coding-agent tokens. An embedded graph with no server removes the cost that kept small teams away from this.

Be clear about the business model. Relate is free, and Aura is the paid step up. That is a fair trade, and openCypher keeps your queries portable. Know it before you start.

What to do:

  1. Read install.sh before you run it. The project page includes a prompt written for AI agents that tells them to pipe the script into bash. Do not let an agent run remote code that you have not read. Pin a release version.
  2. Test it on one multi-step question. Take a query your current agent gets wrong. Load a small data set and compare the answers.
  3. Use file mode for anything you want to keep. In-memory mode loses the data when the process stops.
  4. Limit what relate serve exposes. An MCP endpoint that runs queries can read all the data in that database. Give the agent a database that holds only what it needs.

Key takeaways

  • Neo4j launched Relate on November 11 at NODES 2026 as an open-source embedded graph database for AI agents
  • It is written in Rust, licensed Apache-2.0, and runs in memory, in a directory or in a single file
  • One Cypher statement can combine relationship queries with vector and full-text search
  • An MCP server and a coding-agent skill are included; the speed claims have no published benchmark
  • Read the install script before an agent runs it, and scope the database the MCP endpoint can read

Agent memory is a data-model decision, not a plugin. We build AI agents on data stores you own, with the schema, the access scope and the exit path written down. See how we build agent systems, or tell us the question your agent cannot answer.

Sources: Neo4j, Relate project site.

  • #neo4j
  • #relate
  • #graph-database
  • #ai-agents
  • #mcp
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.