DataVoA C#-native embedded database engine

v0.1 Alpha documentation for allocation-aware SQL, LSM storage, vector search, and local-first .NET workloads.

Native AOT compatible core

DataVo.Core and DataVo.Data enable trim/AOT analyzers and treat IL2026/IL3050-family diagnostics as errors, keeping the engine aligned with Native AOT constraints.

Zero-allocation hot paths

Typed rows, arena-backed LSM MemTables, compiled access paths, and binary WAL frames avoid dictionary materialization on selected internal write and read paths.

Roslyn compiled queries

The DataVo.Generators package emits compiled query plans, typed row mappers, and optional schema-manifest index hints for supported SELECT, INSERT, and UPDATE shapes.

Vector search

Store VECTOR(n) columns, build HNSW or FLAT vector indexes, and rank by cosine (<=>) or L2 (<->) distance directly in DataVo SQL.

DataVo v0.1 Alpha is an attempt to build an embedded database engine from inside the .NET runtime instead of behind a native provider boundary. It is written in C#, runs in-process, and is designed around Span<T>, source-generated serialization, Native AOT constraints, typed row paths, SIMD kernels, and GC-aware execution. In the checked-in benchmark artifacts, DataVo reaches roughly 1.2M+ ops/sec in relaxed LSM thread-scaling workloads, with near-zero or sharply reduced allocation on the hottest internal paths.

This is not a full RDBMS and it is not a mature SQLite replacement. The interesting part of DataVo is narrower: C#-native embedded storage for local-first apps, game tooling, simulations, browser/WASM workflows, reactive query maintenance, and vector search experiments. The current engine supports SQL-style DDL/DML/querying, disk and in-memory modes, WAL/MVCC work, vector columns/indexing, reactive subscriptions, an ADO.NET-facing package, and early EF integration, but the public packaging and production-hardening story is still alpha.

DataVo is being opened alongside the whitepaper because the design tradeoffs are worth discussing in public: how much can a database gain by knowing it lives inside .NET, avoiding avoidable GC pressure, and binding hot paths earlier than a generic provider can? The project is best suited today for early users who like embedded systems, storage engines, .NET performance work, and honest alpha software.

What to read first