DATABASE

14 engines. 1 system.

Stop running Postgres, Redis, Elasticsearch, TimescaleDB, Neo4j, and MongoDB as separate services. Nucleus gives you 14 specialized storage engines in one process — each purpose-built for its data model. Built in Rust with 3,724 passing tests.

3,724

Tests Passing

Data Models

ACID

Transactions

Rust

Built With

Native Multi-Storage Architecture

Nucleus isn't a row-based database with extensions. It's 14 specialized data models in one system, each optimized for its workload.

SQL (Row-Based)

B-tree indexes, MVCC, WAL, PostgreSQL wire protocol. Full OLTP with joins, subqueries, CTEs.

OLAP (Columnar)

Columnar compression, vectorized execution, SIMD scans. 5.9x faster than PostgreSQL for SUM queries.

Vector (HNSW)

Hierarchical Navigable Small World + IVFFlat for approximate nearest neighbor search.

Graph (CSR)

Compressed Sparse Row with index-free adjacency for traversals. Efficient graph storage.

Timeseries

Delta compression, time-based partitioning, downsampling. Purpose-built for metrics and IoT.

Document

JSON collections with B-tree indexing and schema validation. Flexible document storage.

Key-Value

LSM trees with TTL, atomic operations. SQL functions for get, set, delete, increment, and expiry.

Full-Text Search

BM25 ranking, stemming, highlighting, fuzzy matching. Inverted index architecture.

Geo (R-trees)

Spatial indexing for radius, polygon, and nearest-neighbor queries.

Blob Storage

Chunked, content-addressed blob storage with deduplication. Store files alongside structured data.

Streams

Append-only log with consumer groups. Redis-compatible stream semantics for event sourcing.

Datalog

Semi-naive evaluation with SQL integration. Recursive queries and rule-based reasoning.

CDC

Change Data Capture for real-time data pipelines. Track every insert, update, and delete.

PubSub

LISTEN/NOTIFY for real-time event broadcasting. Push updates to connected clients instantly.

Performance Targets

Native storage for each model means Nucleus matches or exceeds specialist databases.

Columnar Engine vs PostgreSQL (measured)

SUM queries

5.9x faster

COUNT(*)

4.2x faster

INSERT

1.6x faster

GROUP BY

1.2x faster

Point query

1.1x faster

Columnar engine benchmarks measured against PostgreSQL. Other engine comparisons in progress.

Database Branching (Planned)

Git for your data. Create isolated branches, test schema changes, then merge back to main. Copy-on-write ensures branches share unchanged data.

main

feature/auth

Test migrations without downtime

Debug production issues in isolation

Staging environments sharing production data

Temporal queries from historical snapshots

Beyond the Engines

Database branching (planned) — Git for your data. Create isolated branches, test schema changes, then merge back. Copy-on-write ensures branches share unchanged data.

Distributed consensus — Raft replication with automatic cluster query forwarding.

Encryption & compression — AES-256 encryption at rest (--encrypt CLI flag). LZ4 compression for storage (--compress CLI flag).

SIMD acceleration — Vectorized AVG/MIN/MAX fast paths for columnar queries. 128MB buffer pool with background flush.

Production-Grade Engineering

3,724 tests passing — unit, integration, and property-based testing with proptest.

Rust 2024 Edition — memory safety without garbage collection. No segfaults, no data races, no null pointer exceptions.

PostgreSQL compatible — use existing Postgres drivers and tools via pgwire 0.36. Drop-in replacement for OLTP workloads.

SQL parsing — powered by sqlparser 0.61 with extensions for multi-model SQL functions.

Comprehensive benchmarks — Criterion benchmarks for query, storage, and specialty workloads. Columnar engine benchmarked against PostgreSQL.

Get Started

$ git clone && cargo build --release$ ./target/release/nucleus --port 5432