Orama — Full-Text and Vector Search Engine for JavaScript

Introduction

Orama is an open-source, dependency-free search engine written in TypeScript that supports full-text, vector, and hybrid search. It runs anywhere JavaScript runs — browsers, Node.js, Deno, Bun, and edge runtimes — providing portable search without external services.

What Orama Does

• Provides sub-millisecond full-text search with built-in typo tolerance and stemming
• Supports vector search and hybrid (keyword + vector) ranking in a single engine
• Runs entirely in-process with no external dependencies or servers required
• Offers faceted search, filters, sorting, grouping, and geo-search
• Includes a plugin system for analytics, persistence, and custom tokenizers

Architecture Overview

Orama stores documents in an in-memory radix tree for text indexing and flat arrays for vector embeddings. Keyword queries traverse the prefix tree while vector lookups use cosine similarity. Results are scored and merged in a single pass with no network overhead. The index serializes to JSON for persistence.

Self-Hosting & Configuration

• Install via npm, yarn, or pnpm with no native dependencies
• Import as an ES module or CommonJS package
• Define a schema object describing document fields and types
• Configure custom tokenizers, stemming, or language settings as needed
• Persist indexes using the persistence plugin or manual JSON serialization

Key Features

• Zero external dependencies; runs on any JavaScript runtime
• Hybrid search combining BM25 full-text ranking with vector similarity
• Built-in typo tolerance with configurable edit distance
• Faceted search, filters, and geo-search out of the box
• Pluggable architecture for extending indexing and search behavior

Comparison with Similar Tools

• Lunr.js — simpler API but lacks vector search, facets, and active maintenance
• FlexSearch — fast full-text but no vector or hybrid capabilities
• MiniSearch — lightweight but limited to keyword search only
• Meilisearch — server-based with broader features but requires a separate process
• Typesense — hosted or self-hosted server; Orama runs in-process without infrastructure

FAQ

Q: Does Orama require a server? A: No. It runs entirely in-process wherever JavaScript runs, including browsers and edge functions.

Q: Can I use Orama with React or Vue? A: Yes. Orama provides framework-agnostic hooks and official integrations for popular frameworks.

Q: How large can the dataset be? A: Browser-side indexes handle tens of thousands of documents; server-side indexes scale to hundreds of thousands.

Q: Does Orama support multiple languages? A: Yes. Built-in stemmers and tokenizers cover 30+ languages.

Sources

• https://github.com/oramasearch/orama
• https://docs.orama.com