NodeDB

What is NodeDB?

NodeDB is a single database engine that aims to replace multiple data systems by combining different data types—relational, vector (AI), graph, document, columnar, and scientific array—into one Rust-based architecture. Its stated goal is to reduce fragmentation between separate databases and remove the need for “pipelines” and “Python glue” when working with mixed data.

A key positioning is that existing PostgreSQL clients can be used “just” to connect, while NodeDB supports GraphRAG-style queries that fuse vector search and graph expansion into one query. The example shown in the page illustrates semantic retrieval plus graph context as part of the database-layer workflow.

Key Features

• Unified engine for multiple data models (relational, vector, graph, document, columnar, scientific array), so you can store and query different data types without switching between separate systems.
• Rust-based architecture, described as a “hyper-efficient” engine implemented in a single Rust binary.
• PostgreSQL client compatibility: the page states that your existing Postgres client “just works,” aiming to lower the friction of adopting a new backend.
• GraphRAG query support that combines vector search with graph expansion in one statement, positioned as “GraphRAG at the database layer.”
• One-query Graph RAG fusion workflow with controls for top-k retrieval, expansion depth, edge label and direction, and result fusion settings (as shown in the example statement).

How to Use NodeDB

1. Get started by joining the early access process or requesting access via the site’s “Get Early Access” option.
2. Use an existing PostgreSQL client to connect, since the page explicitly states Postgres client compatibility.
3. Submit a single GraphRAG-style query that performs semantic retrieval from vectors and expands through graph edges according to the parameters in the query.
4. Use the fused results from that query as the basis for LLM context, since the page frames this as semantic retrieval plus graph context delivered at the database layer.

Use Cases

• Build GraphRAG retrieval without external pipelines: run one database query that performs vector-based semantic retrieval, expands related entities via graph edges, and fuses results for downstream LLM use.
• Answer entity-centric questions with graph context: retrieve top vector matches and then expand through relationships (using edge labels and direction) to gather nearby graph information in the same query.
• Implement hybrid retrieval where ranking and structure matter: use the shown fused retrieval parameters (e.g., top-k, expansion depth, and fusion settings) to balance direct vector matches with graph-expanded results.
• Reduce application-side orchestration: avoid “pipelines” and “Python glue” by moving the fusion of vector and graph operations into the database query itself.
• Consolidate data storage across multiple model types: when an application currently relies on separate systems for relational data, vectors, and graph relations, use NodeDB as a single engine covering those categories.

FAQ

• Does NodeDB require separate pipelines or Python glue to combine vector and graph retrieval? The page states the approach uses “one query” with “No pipelines” and “No Python glue,” describing fusion at the database layer.

• What does “Your existing Postgres client just works” mean? The site explicitly claims PostgreSQL client compatibility, implying you can use common Postgres client patterns to connect.

• What is GraphRAG in this context? The page frames GraphRAG as “vector search + graph expansion” fused in one query, producing semantic retrieval results together with graph context for an LLM.

• What kinds of data models does NodeDB support? The page lists relational, vector, graph, document, columnar, and scientific array data.

Alternatives

• Separate vector database + separate graph database: this keeps vector search and graph traversal in different systems, typically requiring an application layer to orchestrate retrieval and fusion (the page positions NodeDB as avoiding pipelines and glue).
• Hybrid search services with external reranking/fusion: some solutions provide semantic search plus reranking, but the workflow may still require orchestration between retrieval and graph/context expansion steps.
• Traditional SQL databases with extensions for vectors/graphs: you may be able to approximate mixed-model queries using add-ons, but the page’s pitch focuses on a unified engine and one-query fusion across data types.
• GraphRAG implemented in the application layer: instead of performing vector search and graph expansion inside the database, an application can run multiple retrieval steps and then assemble context for the LLM.