Multimodal database for SQL, graph, and vector search

AionDB is a multimodal database, also known as a multi-model database, designed to keep relational data, graph relationships, and vector embeddings in one Rust engine.

The practical goal is to avoid copying the same application data into a SQL database, a graph database, and a vector database when the product needs hybrid queries.

Why a multimodal database?

Modern applications often need several data shapes at once:

• tables for users, documents, tickets, products, events, and business state;
• graph relationships for exploring connections between those records;
• vectors for semantic search, recommendations, RAG, and private AI assistants.

AionDB exposes those models through one PostgreSQL-compatible system. Tables stay the source of truth, while graph and vector queries can operate over the same records.

Hybrid queries

AionDB targets queries that combine SQL filtering, graph traversal, and vector similarity:

MATCH (u:User)-[:WROTE]->(d:Document)-[:CITES]->(ref:Document)
WHERE d.kind = 'runbook'
RETURN d.title,
       ref.title,
       l2_distance(d.embedding, '[0.1,0.8,0.2]') AS distance
ORDER BY distance ASC
LIMIT 5;

This is useful for knowledge bases, product catalogs, support tooling, private copilots, and applications that need both structured business context and semantic retrieval.

PostgreSQL-compatible tooling

AionDB speaks the PostgreSQL wire protocol. The goal is to preserve a familiar developer workflow with psql, pgAdmin, migrations, ORMs, and standard drivers when the required features are compatible.

Project status

AionDB is alpha software. It does not claim to replace PostgreSQL in production today. The narrower claim is that it gives teams an experimental engine for evaluating a multimodal database that combines SQL, graph, and vector search without adding multiple separate services.

Start with the documentation and the reproducible benchmarks.