Skip to content
Projects

hash-anu/snkv has 80 GitHub stars and 3 forks. GitHot measured 0 stars in 24 hours and +1 in 7 days. Last checked 2026-11-15 13:36 UTC. Last push 2026-11-05.

How these counts are measured

What you can do

  • ACID transactions and WAL mode (concurrent readers, single writer)
  • Native TTL: per-key expiry, lazy eviction, bulk kvstore_purge_expired(). No background thread.
  • Encryption: per-value XChaCha20-Poly1305 with Argon2id key derivation, transparent to all APIs
  • Column families: logical namespaces within one database

Measured star growth

in 24h
0
in 7 days
+1
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

SNKV — a lightweight key-value store focused on simplicity, security and performance file based database , written using sqlite's b-tree directly. Now supports vector HNSW , Join the discussion: https://discord.gg/Wwe9dFy3P

Data

README.md

Original

SNKV Logo

SNKV

A simple, crash-safe embedded key-value store

Build Memory Leaks Tests Discord License

SNKV is a lightweight, ACID-compliant embedded key-value store written in C. It talks directly to SQLite's B-tree storage engine and skips the SQL layer: no parser, no query planner, no VM.

SQLite:  Application → SQL Parser → Query Planner → VDBE → B-Tree → Disk
SNKV:    Application → KV API → B-Tree → Disk

Looking for Python? The Python bindings are no longer maintained on master. Use the python_bindings branch.


Quick Start

Single-header integration:

#define SNKV_IMPLEMENTATION
#include "snkv.h"

int main(void) {
    KVStore *db;
    kvstore_open("mydb.db", &db, KVSTORE_JOURNAL_WAL);

    kvstore_put(db, "key", 3, "value", 5);

    void *val; int len;
    kvstore_get(db, "key", 3, &val, &len);
    printf("%.*s\n", len, (char*)val);
    snkv_free(val);

    kvstore_close(db);
}

Features

  • ACID transactions and WAL mode (concurrent readers, single writer)
  • Native TTL: per-key expiry, lazy eviction, bulk kvstore_purge_expired(). No background thread.
  • Encryption: per-value XChaCha20-Poly1305 with Argon2id key derivation, transparent to all APIs
  • Column families: logical namespaces within one database
  • Iterators: forward and reverse, prefix, and seek
  • Vector search: HNSW index via usearch (make vector), stored in the same .db file. See examples/vector.c.
  • Thread-safe: share one handle (one lock per handle; a transaction stays with the thread that began it), or give each thread its own handle for parallel reads in WAL
  • Single-header, Valgrind-clean
  • Compatible with SQLite page/WAL-level tools such as LiteFS. Tools that need SQLite's schema layer (the sqlite3 CLI) do not work.
/* TTL */
kvstore_put_ttl(db, "session", 7, "tok123", 6, kvstore_now_ms() + 60000);

/* Encryption */
kvstore_open_encrypted("secure.db", "hunter2", 7, &db, NULL);

/* Custom configuration */
KVStoreConfig cfg = {0};                  /* zero fields are NOT defaults: set journalMode */
cfg.journalMode = KVSTORE_JOURNAL_WAL;
cfg.syncLevel   = KVSTORE_SYNC_FULL;
cfg.busyTimeout = 5000;
kvstore_open_v2("mydb.db", &db, &cfg);

Build

Linux / macOS

make                  # libsnkv.a
make snkv.h           # single-header version
make examples         # build examples
make test             # run test suite
make vector           # libsnkv_vec.a (core + usearch, requires g++)
make test-vector      # run vector tests

Windows: use the MSYS2 MinGW 64-bit shell (not cmd.exe or PowerShell):

pacman -S --needed mingw-w64-x86_64-gcc make
make && make test

Benchmarks

1M records, mean of 5 runs on Linux (WSL2, tmpfs — measures engine CPU cost, not disk). Both stores use the same SQLite 3.51.2 core built with gcc -O2, WAL with auto-checkpoint every 1000 pages, synchronous=NORMAL, an 8 MB cache and 4 KB pages; SQLite uses a WITHOUT ROWID table and prepared statements. Source: SNKV · SQLite · details in docs/benchmark.html

Benchmark SQLite SNKV Speedup
Sequential writes 1.30M ops/s 1.51M ops/s 1.16x
Random reads 254K ops/s 588K ops/s 2.32x
Sequential scan 11.1M ops/s 13.5M ops/s 1.22x
Random updates 108K ops/s 120K ops/s 1.11x
Random deletes 124K ops/s 126K ops/s 1.01x
Exists checks 262K ops/s 653K ops/s 2.49x
Mixed workload 254K ops/s 317K ops/s 1.25x
Bulk insert 1.50M ops/s 1.75M ops/s 1.17x

Comparisons against LMDB and RocksDB are available too.

Good fit: read-heavy or mixed workloads (point reads and existence checks run about 2.3–2.5x faster than SQLite), native TTL (sessions, caches, leases), embedded or memory-constrained targets, and simple C integration. Consider alternatives: RocksDB for maximum write throughput, LMDB for maximum read speed when memory is plentiful.


Third-Party Licenses

Library License Use
SQLite Public Domain B-tree, pager, WAL, OS layer (bundled)
Monocypher CC0-1.0 XChaCha20-Poly1305 + Argon2id (bundled)
usearch Apache 2.0 HNSW vector index (optional, make vector)

License

Apache License 2.0 © 2025 Hash Anu