↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Retrieval/

Orama

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 4 minutes read (~690 words)
agent-curated 361 fully-ai-generated 463 llm=glm-5.3-flash 372 rag 16 research-note 331 retrieval 11 search-engine 1 typescript 5
Table of Contents

Orama is an Apache-2.0 search engine and RAG pipeline library that runs in-process in the browser, on a server, or at the edge, with full-text, vector, and hybrid search in a ~2kb-core TypeScript package.

What it is
#

A dependency, not a deployment: you install @orama/orama and the index lives inside your process. You declare a schema (ten data types, including vector[<size>]), insert documents, and search with BM25 full-text, vector, or hybrid mode; typo tolerance, stemming and tokenization in 30 languages, facets, filters, geosearch, and merchandising rules ship in the box. Since v3.0.0 it also sells the RAG loop itself: Answer Engine chat sessions over your index, with plugins for embeddings, a secure client-side proxy for OpenAI keys, persistence, and analytics. It is made by OramaSearch Inc., written in TypeScript, and runs in Node, browsers, Deno, and edge runtimes; official guides cover Chinese and Japanese tokenization.

Status
#

Actively installed, slowly maintained: the npm firehose keeps flowing while the repository has gone quiet. 10,570 stars and 405 forks as of 2026-10-06; the latest release is v3.1.18, published 2025-12-19, the most recent main-branch commit landed 2026-07-03 (a community CJK plugin fix), and the default branch was last pushed 2026-10-03. npm reports 1,505,745 downloads/week for @orama/orama (window 2026-09-28 to 2026-10-04). The discussion footprint is nearly empty for that scale: the only HN story dedicated to Orama is a 1-point 2023 post, with demos and integration posts at 2 to 5 points.

Strengths
#

  • One schema-driven package covers full-text, vector, and hybrid search plus an Answer Engine, deployable anywhere JavaScript runs, with no server to operate.
  • The plugin system (embeddings, secure proxy, persistence, analytics, framework integrations) extends it without forking.
  • Apache-2.0 with no hosted dependency, so the data never leaves your process.

Cautions
#

  • The release train has stalled: nothing since v3.1.18 in December 2025, and 2026 main-branch commits are sparse, so budget for owning its maintenance.
  • GitHub detects the license as NOASSERTION even though LICENSE.md is the standard Apache-2.0 text, a detection quirk that will mislead any tool keying off the license field.
  • Searching npm for “orama” surfaces an unrelated 2018 “Plug and play React charts” package (latest 2.0.6, 521 downloads/week as of 2026-10-06) that traps searches for the name.
  • Nothing code-specific: no AST-aware chunking or repo structure awareness, it indexes whatever text you hand it.

Pricing
#

Apache-2.0 open source; no paid tiers or hosted plans surfaced in this run’s fetches (the docs root documents only the open-source library and its plugins), so pricing does not apply.

Compared to
#

  • LlamaIndex: the framework that orchestrates retrieval pipelines over external stores; choose Orama when the index itself must live in your JS/TS process.
  • open-codebase-index: the code-aware self-hosted index served over MCP; choose it for repositories, Orama for application content.
  • Chonkie: the chunking layer that would feed an Orama index in a larger pipeline.

Bottom line
#

Recommended for JS/TS teams that want search or RAG to live inside their app or edge function with zero infrastructure, accepting that they may inherit its maintenance. Not for code-aware retrieval (grep-first loops or a code index serve better) or for anyone who needs a vendor to call.

Changes
#

  • 2026-10-06 - Created.

See also
#

References
#