IndexedDB Adapter

Browser-persistent full-text search using SearchFn's built-in engine with IndexedDB storage.

The IndexedDB Adapter runs SearchFn's built-in search engine in the browser with persistent storage. It uses the same BM25 scoring, inverted index, fuzzy matching, and text pipeline as the Memory Adapter, but stores the inverted index in IndexedDB so search data survives page reloads and browser restarts.

When to Use

Production offline-first browser applications
Progressive web apps with local search
Client-side search that needs to persist across sessions and work without network
Any browser app where you want full-text search without a server round-trip

Installation

npm install @searchfn/adapters

Usage

import { IndexedDbAdapter } from "@searchfn/adapters";

const adapter = new IndexedDbAdapter({
  dbName: "my-app-search",
});

await adapter.initialize({
  resources: [{ name: "tasks", searchFields: ["title", "description"] }],
});

await adapter.index({
  resource: "tasks",
  documents: [
    { id: "t-1", fields: { title: "Ship docs", description: "Write the guide" } },
  ],
});

const ids = await adapter.search({ resource: "tasks", query: "docs" });

Configuration

Option	Type	Default	Description
`dbName`	`string`	`"searchfn-adapter"`	Base database name. Per-resource databases use `{dbName}-{resource}`.
`pipeline`	`PipelineOptions`	Internal defaults	Configure the text pipeline (tokenization, stemming, stop words, n-grams).
`cache.terms`	`number`	`2048`	Maximum entries in the term lookup LRU cache.
`cache.vectors`	`number`	`512`	Maximum entries in the vector LRU cache.

const adapter = new IndexedDbAdapter({
  dbName: "my-app-search",
  cache: { terms: 4096, vectors: 1024 },
});

Capabilities

Capability	Supported
`persistent`	Yes
`searchAll`	Yes
`fuzzy`	Yes
`fieldBoosts`	Yes

How It Works

The IndexedDB Adapter uses the full @searchfn/core search engine with IndexedDB as the storage backend:

Indexing — documents pass through the text pipeline (tokenize, normalize, stop words, stem). Terms and posting lists are written to IndexedDB object stores, with postings stored as delta-varint encoded chunks keyed by [field, term, chunk].
Searching — query text passes through the same pipeline. Postings are retrieved from IndexedDB (with LRU caching for hot terms) and scored using BM25 with length normalization.
Ranking — results are sorted by score (descending) with deterministic tie-breaking by document ID.

LRU Caches

To reduce IndexedDB reads, the adapter maintains two LRU caches:

Term cache (default 2048 entries) — caches frequently accessed posting chunks.
Vector cache (default 512 entries) — caches computed vectors for faster scoring.

Tune these via the cache configuration when working with large indexes.

Behavior

Data persists in IndexedDB across page reloads within the same browser origin.
Each resource gets its own IndexedDB database ({dbName}-{resource}).
Reads and writes are deterministic across repeated calls.
Calling dispose() closes all database handles. Call initialize() again to reopen.

IndexedDB Adapter

On this page