At NODES 2026, Neo4j launched Relate, a new open-source embedded graph database written in Rust (Apache 2.0). It starts and stops with your application, with no server to deploy and no account to create. Agents can store what they learn as a connected graph and use one Cypher statement to follow relationships and run vector and full-text search together. It ships with a coding-agent skill and an MCP endpoint, and switches to a token-saving output mode when it runs under a coding agent. v1.0.1 is out on GitHub, but the README says it is still an experimental Neo4j Labs project outside the supported product line.
Key Takeaways
- ✓Zero deployment: one command installs it and it starts and stops with your app; data lives in memory, a directory or a single .rl file, with a WAL and Parquet checkpoints for durability
- ✓Two kinds of search in one Cypher query: one statement can follow relationships and run vector or full-text search, with no glue code to merge results from two stores
- ✓Compatibility: Neo4j says it runs more than 30,000 openCypher, Cypher 25 and GQL test scenarios to match the Neo4j query surface
- ✓Many languages and entry points: clients for Rust, Python, JS/WASM, Java, Go, Swift and more, plus an optional single-process server that speaks Bolt, HTTP and MCP
- ✓Release status: v1.0.0 shipped on GitHub on Nov 11 and v1.0.1 on Nov 12 under Apache 2.0; no official performance benchmarks yet
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and the problem
Agents juggle documents, tickets, messages and their own notes, and that data is connected: which document a ticket cites, which decision rests on which fact. Traditional databases need a server and an account, which agents tend to avoid, so developers fall back to Python dicts or JSON files that keep the facts but lose the links. A pure vector store finds text that looks similar but cannot answer what depends on what.
Architecture and how it works
Relate is an embedded property-graph engine written from scratch in Rust. It runs as a library inside the application process and starts in milliseconds. Data can live in memory, in a directory or in a single file, with a write-ahead log and Parquet checkpoints for durability. It speaks Cypher, and Neo4j says it runs more than 30,000 openCypher, Cypher 25 and GQL test scenarios. Under a coding agent it switches to an agent mode that returns compact TOON tables, caps rows per query and gives actionable exit codes.
Benchmarks
Neo4j only claims millisecond startup and sub-millisecond queries (vendor figures). It has not published detailed performance numbers or comparisons with other embedded options such as SQLite or Kuzu. For compatibility, every pull request runs Neo4j's Cypher test suites and the coverage is reported publicly.
Getting started
Install with the official relatedb.ai/install.sh script, then run relate skill install to add the skill to coding agents such as Claude Code, Cursor and Codex. You can also use pip install relate-db or npm install @relate/relate. Agents without a shell can connect over MCP through relate serve. Good fits include agent memory, multi-agent task coordination, codebase context and offline on-device apps, and pull/push can sync a graph with Neo4j Aura as a project grows. It is still an experimental Neo4j Labs project, so weigh the risk before using it in production.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.