Everything search should already do.
01 · CRAWL
Knows your site without you helping.
Point us at a domain. We respect robots.txt and sitemaps, render JavaScript, follow redirects, and discover routes you forgot existed. Re-crawls run on a schedule you control — or fire on push via webhook.
- ✓Headless rendering — handles SPAs, deferred content, JS-rendered pages
- ✓Sitemap + robots — sitemap.xml, sitemap_index.xml, robots.txt aware
- ✓Include / exclude patterns — regex per route to scope what gets indexed
- ✓Webhook re-crawl — POST a URL to refresh on deploy
Source page
heading · h2
Configuring filters
paragraph
Filters are configured per metadata field. Define them...
code · html
<script data-filters="true"...
list · 3 items
checkboxes, ranges, deep-link URLs
Embedding · 768d
[0.142, -0.087, 0.213, 0.094,
-0.156, 0.231, ...]
02 · INDEX
Embeddings done right.
Pages are chunked semantically — by heading, list, code block, table — not arbitrarily by character count. Each chunk is embedded with a current-generation model and stored in a hybrid index that takes both vector similarity and lexical signals.
- ✓Semantic chunking — preserves heading and list structure, never mid-paragraph cuts
- ✓Hybrid retrieval — vector similarity + BM25 + reranker, merged transparently
- ✓Multilingual — a query in one language can match content in another
- ✓Edge-served — query path runs in CloudFront's nearest region to the visitor
03 · ANSWER
Sourced AI answers.
Every answer cites the page it came from. We never invent content. If we can't answer from your pages, we say so and offer best-effort search results instead — instead of guessing.
- ✓Inline citations — superscript markers tied to the source URL
- ✓Multi-turn chat — sessions remember context, follow-ups respect it
- ✓Streaming responses — tokens render as they arrive, not all-or-nothing
- ✓Custom system prompt — set tone, rules, refusal behavior per site
Refunds are available within 30 days of purchase, no questions asked[1]. After 30 days, prorated refunds apply for unused billing periods[2]. Annual plans always retain a 7-day full-refund window[3].
Type
Version
Updated
Configuring filters
/docs/v2/filters
Multi-select facet behaviour
/docs/v2/filters/multi
Schema-aware ranges
/docs/v2/filters/range
04 · FACETS
Filter by anything.
Define a metadata schema once. Versions, content types, dates, prices, custom tags — surface them as facets and let users narrow without leaving the page. Filter state is in the URL, so links are shareable and the back button works.
- ✓Schema editor — define facetable fields visually in the dashboard
- ✓Multi-select + ranges — checkboxes, sliders, date pickers, all baked in
- ✓Live counts — facet item counts update as filters narrow
- ✓Deep-linkable — facet state in URL params, browser-history aware
05 · BOOSTS
Your ranking, your call.
Pin specific docs to the top. Weight a URL prefix higher. Tag-boost newer content. Boosts are configured per site and applied transparently — you can see exactly why a result ranks where it ranks.
- ✓Pin to top — guarantee placement for the canonical answer
- ✓URL-prefix boosts — weight
/docs/v2/*higher than/docs/v1/* - ✓Tag boosts — surface anything tagged
recentfirst - ✓Explainable — debug overlay shows which boosts fired
Default
With boosts
--accent: #E0826B --accent: #84CC16 --accent: #111111 Same engine · three sites · zero shared CSS
06 · THEMING
Looks like you, not us.
CSS variables for everything: colours, spacing, fonts, radii, shadows, animation timing. Slot in your own React or web components for total control. Light and dark modes ship in parity.
- ✓22 CSS variables — type, colour, spacing, radii, shadow, motion timing
- ✓Component slots — bring your own result, citation, empty-state UI
- ✓Light + dark parity — the widget mirrors your site's mode automatically
- ✓Headless mode — drop the styles entirely, ship your own
The boring-but-vital stuff, also done.
Analytics, multi-language, webhooks, security, scale. The plumbing nobody puts on a marketing page — but the things that decide whether you'll still trust us in eighteen months.
Analytics
Top queries, zero-result hits, click-through. Per-site, per-mode, viewable in the dashboard.
Multilingual
A French query finds an English page if it answers the question. The model handles the bridge.
REST · MCP · OpenAPI
OpenAPI 3.0 spec. Hosted MCP server for agents. Webhook re-crawl. Build whatever you want on top.
Managed, encrypted hosting
Runs on Quant Cloud — multi-region, encrypted at rest. Prompts are never retained or used to train models; region pinning on request.
Running Drupal? Skip the script tag.
There's a turnkey module. One-click OAuth from your Drupal admin, real-time content indexing as nodes are saved, drop-in widget blocks via the standard Block Layout. Drupal 10 or 11.
- ✓OAuth connect — authenticate to QuantSearch from
/admin/config/search/quantsearch, no API keys to copy - ✓Real-time indexing — node save / publish / delete events sync to the index automatically
- ✓Drush commands —
drush qs-indexfor bulk operations,qs-crawl,qs-purge - ✓Hookable —
hook_quantsearch_ai_page_alter()lets you tweak the payload before send
$ composer require quantcdn/quantsearch-ai
$ drush en quantsearch_ai
→ module enabled · Key dependency installed
→ visit /admin/config/search/quantsearch
✓ click "Connect to QuantSearch" → done
The questions we get most.
Why not just send visitors to ChatGPT? +
Generic LLMs are trained on the open web, often months out of date, and will confidently invent answers about your product. QuantSearch only answers from your own indexed pages, with an inline citation on every claim. Visitors get authoritative answers tied to the docs you wrote — and they stay on your site instead of bouncing to a search engine.
Where does my data live? +
Compute and the search index run in AWS regions selected for low-latency delivery. Usage logs are kept separately for billing only. AI inference runs on AWS — prompts and responses aren’t retained or used to train models. We can pin a region for your tenant on request; the default is the closest available to your users.
Will the index stay in sync as our content changes? +
Schedule re-crawls daily, hourly, or whatever rhythm fits. POST to a webhook from your deploy pipeline and the index refreshes on push. Running Drupal? The module syncs nodes in real time as they're saved — publish a page and it's searchable seconds later.
Does it work for non-English content? +
Yes — our embedding model is multilingual and covers 100+ languages. A French query finds an English page (and vice versa) if it answers the question. Useful for multi-locale docs sites and any audience that asks in their own language.
How do I plug this into Claude, Cursor, or our own agents? +
There's a hosted MCP server at mcp.quantsearch.ai. Add one line to your agent's config and the indexed corpus becomes a callable tool. No install, no key rotation — just a URL and a site identifier. Anything that speaks MCP can search and chat against your content.
Can I make it look like our brand? +
22 CSS variables cover colours, fonts, radii, shadows, and motion timing — drop them in your stylesheet and the widget inherits. For deeper customisation, slot in your own React or web components for results, citations, and empty states. Light and dark modes ship at parity automatically, mirroring whatever your site does.
Can we deploy this inside our own AWS account? +
Out of the box, QuantSearch is fully managed — it's how most customers want it. For specific cases (existing AWS Organization, data sovereignty rules, air-gapped environments), we'll scope a deployment into your own account on a case-by-case basis. Talk to sales →
Your site has the answer.
Help them find it.
Free tier, no credit card. Cancel by removing the script tag — that's the whole thing.
Setup
One script tag
Uptime
100% since launch
Hosting
Multi-region
Cancel
Delete the tag