SQLite Store

The store/sqlite package provides a SQLite-backed MetadataStore using the Grove ORM with sqlitedriver. It is suitable for embedded applications, CLI tools, and single-node services that don't require a separate database server.

Installation

go get github.com/xraph/weave/store/sqlite

Setup

import (
    "github.com/xraph/grove"
    "github.com/xraph/grove/drivers/sqlitedriver"
    sqlitestore "github.com/xraph/weave/store/sqlite"
)

// Open a SQLite connection via Grove's sqlitedriver.
db := grove.Open(sqlitedriver.New("weave.db"))

store := sqlitestore.New(db)

Migrations

if err := store.Migrate(ctx); err != nil {
    log.Fatal("migrate:", err)
}

Creates the same three tables as the PostgreSQL store (weave_collections, weave_documents, weave_chunks) using the Grove migration orchestrator with SQLite-compatible column types.

Pairing with a vector store

SQLite does not support pgvector. Pair it with the in-memory vector store for single-node use:

import (
    sqlitestore "github.com/xraph/weave/store/sqlite"
    memvec "github.com/xraph/weave/vectorstore/memory"
)

eng, err := engine.New(
    engine.WithStore(sqlitestore.New(db)),
    engine.WithVectorStore(memvec.New()),
    engine.WithEmbedder(emb),
)

Internals

Lifecycle

When to use

• Local development -- single-file database, no server process required.
• Testing -- fast, isolated store with no external dependencies.
• Embedded / single-process -- CLI tools, desktop apps, or edge deployments.
• CI pipelines -- deterministic tests without a database container.

Limitations

• Single-writer -- SQLite uses database-level write locks. High write concurrency will serialize.
• No pgvector -- vector similarity search requires a separate vector store.
• Not horizontally scalable -- suited for single-process or embedded use.