One Core.
Every Model.
A converged multimodal database engine unifying Relational SQL, AI Vector RAG, NoSQL JSON Documents, and PL/SQL Scripting into a single storage engine with zero external dependencies.
In-Browser Multimodal Engine
Run real Relational Queries, HNSW Vector Distance, JSON Traversal, and PL/SQL Procedural Scripts compiled directly into JavaScript via Dart WebAssembly.
|
Click Run Query above to execute against the pure Dart engine.
|
Four Paradigms. One Engine.
Unified storage engine executing relational queries, high-dimensional vector embeddings, document queries, and stored procedural scripts.
Relational SQL Core
Complete ACID compliance with MVCC snapshot isolation, multi-table joins, nested subqueries, and B-Tree indexing on 4KB Slotted Pages with CRC32 verification.
- ANSI-SQL JOIN, GROUP BY, HAVING
- Write-Ahead Logging (WAL) Recovery
- Multi-Column Unique & Foreign Keys
FROM users u INNER JOIN orders o
ON u.id = o.user_id GROUP BY u.name;
AI Vector RAG (HNSW)
Native Hierarchical Navigable Small World (HNSW) graphs integrated directly into the database engine. Execute millisecond vector search over high-dimensional embeddings.
- Cosine, Euclidean & Dot Product distance
- 384, 768, 1536, and 3072 dimensions
- Hybrid filtered search with SQL predicates
FROM documents WHERE status = 'active'
ORDER BY 2 ASC LIMIT 5;
NoSQL JSON Documents
Store schema-less nested JSON documents with native dotted path queries, array indexing, and automatic validation without schema lock-in.
- Direct extraction via -> and ->> operators
- In-place nested attribute mutations
- Seamless combination with Relational tables
FROM profiles
WHERE data->>'plan' = 'pro';
PL/SQL Scripting
Execute complex procedural logic directly inside the database kernel. Declare typed variables, loops, conditionals, and atomic transactions without network round-trips.
- DECLARE, BEGIN, EXCEPTION, END blocks
- WHILE / FOR loops & IF-THEN branching
- Transactional rollback on script error
BEGIN WHILE cnt < 10 LOOP
cnt := cnt + 1; INSERT INTO t VALUES (cnt);
END LOOP; END;
Engine Throughput & Latency
Audited performance under automated workloads against native C SQLite, Dart Drift FFI, and Hive.
Engineered for Reliability & Portability
Designed from the ground up for modern Dart & Flutter applications across Mobile, Desktop, Server, and Web.
Zero Native C/C++ Dependencies
No CMake, NDK, Xcode toolchain mismatches, or linking headaches. In Python and Node.js, a small bundled engine process starts automatically.
4KB Slotted Pages with CRC32
Slotted page file format with variable-length tuple offsets and hardware-accelerated CRC32 checksums on every disk read.
Write-Ahead Logging (WAL)
Guarantees durability with ARIES-style redo logging and atomic transaction commit markers. Instant recovery on unexpected app termination.
PostgreSQL (pgwire) Compatible
Embedded Pgwire server allowing standard clients like psql, DBeaver, and Supabase studio to connect directly to your ULTSQL database.
MVCC Snapshot & Shared Read Locks
Multi-Version Concurrency Control plus shared file lock escalation (`readOnly: true`), allowing multiple processes to execute concurrent reads without lock contention.
Single-File (.ultsql) & In-Memory
Zero external server dependencies. Run completely ephemeral in RAM or persist transparently to a clean single `.ultsql` file on disk.
Trigram Indexes (USING TRIGRAM)
Accelerate `LIKE '%substring%'` searches with inverted 3-gram posting lists and set intersections, avoiding full table page scans.
Raft Consensus & Delta-CRDT
Production-grade multi-node Raft consensus clustering with quorum log replication, alongside Hybrid Logical Clock (HLC) Delta-CRDT offline synchronization.
Opt-in SDKs & 'ultsql dev'
Python & Node.js client SDKs with opt-in `auto_start`, loopback security (`127.0.0.1`), process reuse, and `ultsql dev` for live local development servers.
Up and Running in 30 Seconds
Add the package, open an instance, and start executing multimodal queries immediately.
dart pub add ultsql
import 'package:ultsql/ultsql.dart';
void main() async {
// 1. Open a single-file database instance
final db = await UltSqlDatabase.openFile('app_data.ultsql');
// 2. Execute multimodal DDL and Vector search
await db.execute('''
CREATE TABLE docs (id INT PRIMARY KEY, content TEXT, embedding VECTOR);
INSERT INTO docs VALUES (1, 'Hello World', '[0.1, 0.9, -0.4]');
''');
// 3. Query with nearest vector distance
final results = await db.query(
'SELECT content, VECTOR_DISTANCE(embedding, :q) AS dist FROM docs ORDER BY dist ASC',
params: {'q': '[0.12, 0.88, -0.42]'},
);
print(results);
}