Zero-Config Client-Side Search Engine

Reef Engine & Agentic Automation

High-performance, in-memory inverted index search and browser agent primitives for static sites. Zero server costs, zero accounts, and zero build steps required.

$0.00

Cloud Cost

100% Client Memory

38ms

Median Latency

BM25F Scoring

~90B

Gzipped Size

Zero Dependencies

100%

Private Data

Zero Telemetry

Zero-Config Script Embed

<script src="https://reef.js.org/dist/reef.min.js" data-actions-mode="execute" data-sitemap="/sitemap.xml" data-hotkey="cmdk" async defer></script>

BM25F Inverted Index

Multi-field BM25F scoring with per-field weights across headingText, bodyText, label, and breadcrumb. Supports MMR diversity re-ranking and popularity-based query boosting.

Agent & Workflow API

Chainable agent() primitives for LLM automation: click, type, submit, navigate, extract, and multi-step JSON/YAML executeWorkflow() with retry logic.

IndexedDB Cache

Persistent IndexedDB caching with ETags, lastmod, and content-hash invalidation. Worker-based indexing offloads crawling from the main thread on supporting browsers.

Zero Telemetry

100% private local execution. Queries, index data, and browsing activity never leave the browser. Shadow DOM UI keeps styles fully isolated from the host page.

Universal DOM Grounding

Semantic selector candidates, ARIA accessibility extraction, fuzzy live-text fallback, open Shadow DOM, and same-origin iframe traversal keep agent actions grounded after re-renders.

Live Observation & Guardrails

observe(), waitForStable(), SPA route awareness, action verification, pagination limits, rate limiting, and destructive-action protection support bounded autonomous exploration.

Site Graph & LLM Tools

crawlAndBuildGraph() maps pages and effects for multi-step planning, while schema-described agentTools connects Reef to tool-use loops.

Agent-Ready Sites

Drop in reef-agent-ready.js to publish a live, guarded manifest that external browser agents can consume without adopting the full Reef UI.

SYSTEM_ARCHITECTURE_PIPELINE

Client-Side Processing Pipeline

Engine: BM25F + Inverted Memory

Reef operates completely inside the browser runtime without sending HTML or query data to backend servers. Click on each phase below to inspect the pipeline logic:

PHASE 01 // INDEXER

Sitemap Discovery & Crawling

The Indexer class (src/indexing/indexer.ts) reads data-sitemap (default: /sitemap.xml) and parses <loc> + <lastmod> entries. Falls back to same-origin breadth-first crawl if no sitemap is available. Pages are fetched as plain text and parsed with DOMParser — no scripts execute.

Execution Specifications:

Parses sitemap.xml <loc> entries; follows one level of sitemap index nesting
Fallback same-origin crawl capped by data-max-pages (default 500)
Respects crawlDelay, honors rel="nofollow", excludes same-page anchors
Worker-based indexing off main thread when data-use-worker-indexing="true"

Core Implementation Fragment:

// Phase 1: Sitemap Discovery (src/indexing/indexer.ts)
const urls = await this.fetchSitemapUrls();
// Falls back to breadth-first crawl if sitemap returns 404
for (const url of urls) {
  const html = await fetch(url).then(r => r.text());
  const doc = new DOMParser().parseFromString(html, 'text/html');
  await extractSections(doc, url, this.index);
}

SOURCE_MODULE_MAP

src/ — TypeScript Source Files

11 modules

browser.ts ENTRY

Instantiates ReefSearch and exposes it as window.Reef. The only file that touches the global scope.

reef.ts ORCHESTRATOR

Top-level ReefSearch class. Coordinates boot, hotkey registration, result rendering, and exposes the full public API (open, search, act, agent()…).

types.ts TYPES

Shared interfaces: IndexRecord, SearchOptions, ReefConfig, WorkflowStep, AgentSession, ScoredRecord, and token filter types.

search-index.ts SEARCH ENGINE

In-memory inverted index with BM25F per-field scoring, MMR diversity re-ranking, Levenshtein fuzzy fallback, prefix trie for suggest(), LRU query cache, and index serialization.

extraction.ts EXTRACTION

Heading-based section splitting plus stable semantic selector candidates, accessibility-tree extraction, shadow-root traversal, iframe paths, JSON-LD parsing, and hidden-content expansion.

cache.ts CACHE

IndexedDB persistence with optional gzip CompressionStream. Stores versionHash, buildTime, per-page ETag/lastModified/contentHash, and the serialized index payload.

agent.ts AGENT

Fluent Agent with live .observe(), .waitForStable(), visibility scrolling, SPA route awareness, selector candidates, fuzzy fallback, verification, bounded pagination, and guardrails.

workflow.ts WORKFLOW

YAML workflow parser, step validator, and executeWorkflow() runner with per-step retry, onStepStart / onStepComplete callbacks, and stopOnError control.

worker.ts WEB WORKER

Web Worker shim handling createIndex, indexPages, search, serializeIndex, and deserializeIndex messages off the main thread via DOMParser.

indexing/indexer.ts CRAWLER

Indexer class: sitemap fetch, recursive sitemap-index processing, parallel page crawl with incremental ETag diffing, prebuilt index URL support, and same-origin fallback crawl.

actions/action-executor.ts EXECUTOR

ActionExecutor: resolves IndexRecord selectors to live DOM elements, dispatches synthetic events, handles same-page scroll-highlight, deferred cross-page scroll via sessionStorage.

config/config-reader.ts CONFIG

ConfigReader parses data-* attributes from the install <script> tag into a typed ReefConfig. ConfigApplier writes CSS custom properties to the Shadow DOM host.

ui/renderer.ts UI RENDERER

UIRenderer: mounts Shadow DOM, renders result rows with match highlighting, manages category tab state, focus trap, aria-live announcements, settings panel, and toast notifications.

ui/inspector.ts DEV TOOL

VisualInspector: overlay that draws dashed bounding boxes over every indexed action (pink) and field (cyan) on the live page, with hover badges showing record IDs.

UNIVERSAL_INDEX_MODEL

IndexRecord Types & Extraction Sources

section

Heading-bounded page content. Each <h1>–<h6> produces one record with headingText, bodyText, breadcrumb, and anchor URL. Primary navigation target.

action

Buttons and clickable controls indexed with a generated CSS selector. Gated by actionsMode: destructive verbs excluded in navigate-only mode.

field

Form inputs, textareas, and selects. Selecting a field record focuses the element. fillField(id, value) uses native property setters for React/Vue compatibility.

file

Downloadable links classified by extension: pdf, doc, xls, ppt, zip, csv, and more (configurable via data-file-extensions).

media

Images, audio, and video elements indexed with alt text, captions, and transcripts as searchable body content.

structured

Parsed from <script type="application/ld+json">. FAQ entries, breadcrumb lists, and other schema.org records become searchable answer-type results.

link

External and internal hyperlinks that are not file targets. Indexed with link text and target URL for navigation-only result entries.

AUTONOMOUS_EXPLORATION_LAYER

Plan across pages and states

Reef can observe the composed DOM, wait for JavaScript-heavy interfaces to settle, detect SPA route changes, and build a bounded SiteGraph of links, fields, actions, and effects. Exploration respects actionsMode, destructive flags, and per-run action limits.

const agent = new Agent(index, inspector, { maxActionsPerRun: 25 });
const actions = await agent.observe();
await agent.waitForStable({ quietMs: 300 });
const graph = await crawlAndBuildGraph(location.href, { maxPages: 20, persist: true });

PROGRAMMATIC_BROWSER_AUTOMATION

Searchable Agentic API

Search by method name, capability, or intent. Select a call to see its real signature, what it changes, and a copyable example. The sandbox below previews the call shape against Reef’s browser-local runtime.

window.Reef primitives

Method calls & capabilities

Search method calls

METHOD	SIGNATURE	DESCRIPTION	EXAMPLE	SAFETY

Method Call Preview

Choose a method call:

Example input:

ONE_SCRIPT_AGENT_INSTRUMENTATION

Agent-Ready Sites

Make any site discoverable to browser agents with a standalone script. Reef stamps durable targets, fills in accessible names for icon controls, traverses open Shadow DOM and same-origin frames, and republishes a fresh manifest after SPA navigation.

GET_STARTED

<script src="https://reef.js.org/dist/reef-agent-ready.min.js" data-mode="agent-ready" async></script>

For a static server-side copy, run npm run export-manifest -- https://your-site.example dist/.well-known/agent-manifest.json with Playwright installed, then deploy it at /.well-known/agent-manifest.json.

Configuration attributes

ATTRIBUTE	DEFAULT	DESCRIPTION
`data-exclude`	empty	Comma-separated CSS selectors excluded from the manifest.
`data-mode`	`execute`	`navigate-only` marks actions destructive for conservative agents.
`data-disable-aria-backfill`	`false`	Disable labels added to unnamed icon controls.
`data-disable-live-updates`	`false`	Disable mutation and route-change rescans.
`data-debounce-ms`	`150`	Debounce interval for live rescans.
`data-max-rescans-per-minute`	`30`	Rate limit protecting pages from runaway updates.
`data-debug`	`false`	Log each published manifest to the console.
`data-publish-well-known`	`false`	Enable deployment guidance for exporting the manifest to `/.well-known/agent-manifest.json`.

HOW_AGENTS_CONSUME_THIS

const manifest = window.__reefAgentManifest;
document.addEventListener('reef:agent-ready', event => {
  const records = event.detail.records;
});
const json = document.querySelector('script[type="application/agent-manifest+json"]');
const serverReady = JSON.parse(json.textContent);

GUARDRAILS: password, payment-card, SSN/social-security, data-sensitive, and data-reef-agent="off" regions are excluded by default. Add site-specific selectors with data-exclude="#admin, .private-widget". The manifest never submits forms or executes actions.

BENCHMARK_MATRIX_SPEC

Architectural Comparison Matrix

Client Inverted vs Server SaaS

Toggle Engines to Compare:

FEATURE_COMPARISON_MATRIX

Search Engine Feature Comparison

14 Tools Analyzed

Comprehensive feature matrix comparing Reef against popular client-side libraries and hosted search services. Green indicates full support, yellow indicates partial support or limitations, and red indicates missing or unavailable features.

Tool	Type	Backend	Cost	Privacy	Live DOM	Fuzzy Search	Indexed	DOM Sync	Type System	BM25	Suggest	Stemming	Faceting	Worker	Serialization
Reef PINNED	Client-side lib	None	Free · MIT	Stays local	Yes — live DOM	Yes (indexed)	Yes (inverted)	Planned	Yes — 7 types	Yes (option)	Yes	Yes (opt-in)	Yes	Planned	Yes
Fuse.js	Client-side lib	None	Free · MIT	Local	No — generic data	Yes (Bitap)	No — full scan	No	No	No	No	No	No	No	Yes
MiniSearch	Client-side lib	None	Free · MIT	Local	No	Yes	Yes (inverted)	No	No	Yes	Yes	Yes	Manual	No	Yes
uFuzzy	Client-side lib	None	Free · MIT	Local	No	Yes (staged)	No — full scan	No	No	No	No	No	No	No	No
FlexSearch	Client-side lib	None	Free · MIT	Local	No	Yes	Yes	No	No	Yes	Partial	Yes	No	Yes	Yes
Lunr.js	Client-side lib	None	Free · MIT	Local	No	Yes (~)	Yes	No	No	Yes	No	Yes	No	No	Yes
Elasticlunr.js	Client-side lib	None	Free · MIT	Local	No	Yes	Yes	No	No	Yes (BM25)	No	Yes	No	No	Yes
fuzzysort	Client-side lib	None	Free · MIT	Local	No	Yes	No (cached scan)	No	No	No	No	No	No	No	No
Orama	Client-side / node	None (in-browser)	Free · MIT	Local	No	Yes	Yes	No	Typed schema	Yes	Partial	Yes (plugin)	Yes	Yes (wasm)	Yes
Pagefind	Static (build-time)	None (static)	Free · MIT	Local shards	No (prebuilt)	Yes	Yes (shards)	No	No	Partial	Partial	Yes	Yes	Yes (wasm)	Yes (shards)
Algolia	Hosted service	Required	Paid (free tier)	Leaves browser	No	Yes	Yes	No	No	Yes	Yes	Yes	Yes	n/a	No (backend)
Meilisearch	Self-host / cloud	Required	Free self-host	To your server	No	Yes	Yes	No	No	Yes	Yes	Yes	Yes	n/a	No (backend)
Elasticsearch/OpenSearch	Self-hosted engine	Required	Free (Apache)	To your server	No	Yes	Yes	No	No	Yes	Yes	Yes	Yes	n/a	No (backend)
Fuse Cloud	Hosted service	Required	Paid	Leaves browser	No	Yes	No — full scan	No	No	No	No	No	No	No	Yes

Legend:

Sample Full support or ideal

Sample Partial or limited

Sample Not available or drawback

SYSTEM_INSTALLATION_EMBED

Installation & Embed Configurator

Single Script / NPM Package

AGENT_ONLY_INSTALL

Make your site agent-ready without search

Paste this standalone script before </body>. It publishes the live agent manifest and does not install Reef’s search UI or search features.

<script src="https://reef.js.org/dist/reef-agent-ready.min.js" data-mode="agent-ready" async defer></script>

Use the agent-ready only option below to copy this exact snippet.

Custom Script Embed Generator

Install mode:

Agent-ready mode omits Reef search. Use it only when this page needs manifest instrumentation and no search UI.

data-actions-mode:

data-sitemap:

data-hotkey:

Paste Before </body> in HTML:

<script src="https://reef.js.org/dist/reef.min.js" data-actions-mode="execute" data-sitemap="/sitemap.xml" data-hotkey="cmdk" async defer></script>

NPM / Package Manager

npm install reef-search

Direct CDN Download GitHub Release

https://reef.js.org/dist/reef.min.js

TECHNICAL_FAQ_INDEX

Frequently Asked Questions

Developer Specification Q&A