KipData/KiteSQL: Embedded relational database and native Rust data API. · GitHub

KiteSQL is a lightweight embedded relational database for Rust, inspired by MyRocks and SQLite and fully written in Rust. It is designed to work not only as a SQL engine, but also as a Rust-native data API that can be embedded directly into applications without relying on external services or heavyweight infrastructure.

KiteSQL supports direct SQL execution, typed ORM models, schema migration, and builder-style queries, so you can combine relational power with an API surface that feels natural in Rust. On native targets, KiteSQL ships with both RocksDB-backed and LMDB-backed persistent storage builders, plus an in-memory builder for tests and temporary workloads.

• A lightweight embedded SQL database fully rewritten in Rust
• A Rust-native relational API alongside direct SQL execution
• Typed ORM models with migrations and a lightweight typed query/mutation builder
• Higher write speed with an application-friendly embedding model
• All metadata and actual data in KV storage, with no intermediate stateful service layer
• Extensible storage integration for customized workloads
• Supports most of the SQL 2016 syntax
• Ships a WebAssembly build for JavaScript runtimes

KiteSQL includes a built-in ORM behind the orm feature flag. With #[derive(Model)], you can define typed models and get tuple mapping, schema creation, migration support, projections, set queries, and builder-style query/mutation workflows.

Model changes are part of the normal workflow. KiteSQL ORM can help evolve tables for common schema updates, including adding, dropping, renaming, and changing columns, so many migrations can stay close to the Rust model definition instead of being managed as hand-written SQL.

For the full ORM guide, see src/orm/README.md.

use kite_sql::db::DataBaseBuilder;
use kite_sql::errors::DatabaseError;
use kite_sql::orm::OrmQueryResultExt;
use kite_sql::Model;

#[derive(Default, Debug, PartialEq, Model)]
#[model(table = "users")]
#[model(index(name = "users_name_age_idx", columns = "name, age"))]
struct User {
    #[model(primary_key)]
    id: i32,
    #[model(unique, varchar = 128)]
    email: String,
    #[model(rename = "user_name", varchar = 64)]
    name: String,
    #[model(default = "18", index)]
    age: Option<i32>,
}

fn main() -> Result<(), DatabaseError> {
    let mut database = DataBaseBuilder::path("./data").build_rocksdb()?;
    // Or: let database = DataBaseBuilder::path("./data").build_lmdb()?;

    database.migrate::<User>()?;

    database.insert_many([
        User {
            id: 1,
            email: "alice@example.com".to_string(),
            name: "Alice".to_string(),
            age: Some(18),
        },
        User {
            id: 2,
            email: "bob@example.com".to_string(),
            name: "Bob".to_string(),
            age: Some(24),
        },
    ])?;

    database
        .bind(|ctx| {
            ctx.mutate::<User>()?
                .filter(|e| e.column(User::id())?.eq(1))?
                .update(|u| u.set_value(User::age(), Some(19)))
        })?
        .done()?;

    database
        .bind(|ctx| {
            ctx.mutate::<User>()?
                .filter(|e| e.column(User::id())?.eq(2))?
                .delete()
        })?
        .done()?;

    let users = database
        .bind(|ctx| {
            ctx.from::<User>()?
                .filter(|e| e.column(User::age())?.gte(18))?
                .project_scalars((User::id(), User::name()))?
                .order_by(User::name())?
                .limit(10)?
                .finish()
        })?
        .project_tuple::<(i32, String)>();

    for user in users {
        println!("{:?}", user?);
    }

    // For ad-hoc or more SQL-shaped workloads, `run(...)` is still available.

    Ok(())
}

• build_rocksdb() opens a persistent RocksDB-backed database.
• build_lmdb() opens a persistent LMDB-backed database.
• build_in_memory() opens an in-memory database for tests, examples, and temporary workloads.
• build_optimistic() is available on native targets when you specifically want optimistic transactions on top of RocksDB.
• Database::checkpoint(path) creates a local consistent snapshot when the selected storage backend supports it.
• Transaction isolation is documented in docs/transaction-isolation.md.
• Cargo features:
  • rocksdb is enabled by default
  • parser is enabled by default and provides the SQL parser frontend
  • spill optionally enables spill-backed external sorting and aggregation
  • spill and wasm are mutually exclusive
  • lmdb is optional
  • unsafe_txdb_checkpoint enables experimental checkpoint support for RocksDB TransactionDB
  • cargo check --no-default-features --features lmdb builds an LMDB-only native configuration

On native targets, LMDB shines when reads dominate, while RocksDB is usually the stronger choice when writes do. Checkpoint support and feature-gating details are documented in docs/features.md.

• Run cargo run --bin kitesql-shell to open the local interactive shell.
• Use cargo run --bin kitesql-shell -- --path ./tmp/kitesql-shell-data to point to a custom RocksDB directory.
• Use cargo run --bin kitesql-shell -- -e "select current_timestamp" for a quick one-shot check.
• In interactive mode, end SQL statements with ;; an empty line also executes the buffered statement.
• Supported metacommands include .help, .quit, .tables, .views, and .schema <name>.

👉more examples

• hello_world
• transaction

• Build: wasm-pack build --release --target nodejs (outputs to ./pkg; use --target web or --target bundler for browser/bundler setups).
• Usage:

import { WasmDatabase } from "./pkg/kite_sql.js";

const db = new WasmDatabase();
await db.ddl("create table demo(id int primary key, v int)");
await db.execute("insert into demo values (1, 2), (2, 4)");
const rows = db.run("select * from demo").rows();
console.log(rows.map((r) => r.values.map((v) => v.Int32 ?? v)));

• In Node.js, provide a small localStorage shim if you enable statistics-related features (see examples/wasm_index_usage.test.mjs).

• Enable bindings with Cargo feature python.
• Constructor is explicit: Database(path, backend="rocksdb"); use backend="lmdb" to open LMDB. In-memory usage is Database.in_memory().
• Minimal usage:

import kite_sql

db = kite_sql.Database.in_memory()
db.execute("create table demo(id int primary key, v int)")
db.execute("insert into demo values (1, 2), (2, 4)")
for row in db.run("select * from demo"):
    print(row["values"])

Run make tpcc (or cargo run -p tpcc --release) to execute the benchmark against the default KiteSQL storage. Use --backend rocksdb or --backend lmdb to compare the two persistent backends directly.
Run make tpcc-dual to mirror every TPCC statement to an in-memory SQLite database alongside KiteSQL and assert the two engines return identical results; this target runs for 60 seconds (--measure-time 60). Use cargo run -p tpcc --release -- --backend dual --measure-time <secs> for a custom duration.

• i9-13900HX
• 32.0 GB
• KIOXIA-EXCERIA PLUS G3 SSD
• Tips: TPC-C currently only supports single thread

Recent stable-run 720-second local comparison on the machine above:

These rows are from the stable runs on 2026-07-11; the detailed raw outputs are recorded in tpcc/README.md.

• Get SQL 2016 mostly supported
• LLVM JIT: Perf: TPCC

KiteSQL uses the Apache 2.0 license to strike a balance between open contributions and allowing you to use the software however you want.