Refleks

Refleks documentation

Documentation with clear boundaries

Refleks connects a public website to an AI assistant. These guides describe the behavior currently implemented in the product, including the places where a feature depends on your site or deployment configuration.

Start here

Connect, inspect, then publish.

Add a public URL in the dashboard, let the crawl finish, test answers from the indexed pages, and install the generated widget snippet only when it looks right. The installation guide walks through the full path.

What is supported?

The labels below are deliberately narrow. A supported path can still be limited by your plan, robots rules, network availability, or the content your site returns.

Supported

Public website pages

HTTP(S) pages on the same hostname as the URL you add, discovered from the start page, same-host links, and same-host sitemaps.
Conditional

JavaScript-rendered content

A configured Crawl4AI browser sidecar can render pages before extraction. Without it, Refleks uses a direct HTTP response and HTML parsing, so client-only content may be absent.
Not supported

External or private targets

External links, private/internal addresses, non-HTTP schemes, non-default ports, and authenticated crawl sessions are not part of the built-in crawl flow.

Install the widget

Add a site, copy the generated script, publish it globally, and use dashboard verification to check the output.

Read guide

Crawling and indexing

Learn how robots.txt, sitemaps, links, canonical URLs, exclusions, dynamic pages, and languages affect the index.

Read guide

Configure the widget

Review the supported appearance fields, secure origins, CSP allowances, and the difference between install and domain verification.

Read guide

Plans and limits

Compare page, site, message, and daily refresh caps—and see exactly what is blocked when a cap is reached.

Read guide

Troubleshooting

Use the crawl progress and skipped-page report to separate a site problem, a scope rule, a quota, and an indexing failure.

Read guide

The product flow

Each stage has a different status and a different failure mode.

  1. 01

    Connect

    Add a URL after email verification. The server checks the current account site allocation.

  2. 02

    Discover

    Robots rules, sitemaps, the seed URL, and same-host links build a bounded queue.

  3. 03

    Index

    Readable content is chunked and embedded. Progress moves from fetching to indexing.

  4. 04

    Answer

    Retrieval supplies page context and source URLs; the model is instructed not to invent missing facts.

Accuracy and citations

Sources improve reviewability, but they do not turn an AI response into a guarantee.

  • Answers are generated from the active indexed snapshot for the site. A page that was skipped, blocked, not yet crawled, or changed after the last successful crawl cannot support a current answer.
  • When retrieval returns supporting chunks, the API sends their page URLs to the widget. A source link can be absent when no relevant source is available, and a link is not proof that the generated wording is correct.
  • The answer prompt tells the model to use only indexed website content and to say it is unsure when the answer is not present. It cannot guarantee perfect extraction, retrieval, or model output.
  • For prices, policies, dates, and other consequential information, review the linked page and keep the website itself current.

For the exact selection and extraction rules, see page selection and accuracy limits.

FAQ

Short answers, with links to the detailed behavior.

Does Refleks crawl my whole domain?

It crawls a bounded set of same-host URLs up to the active plan’s page limit. Sitemaps and links influence discovery; external hosts, skipped file types, robots-blocked URLs, and short pages are excluded. See Crawling.

Can I use a subdomain?

Yes, when you add that subdomain as the starting hostname. A crawl started at docs.example.com does not automatically include www.example.com or example.com.

Is there a 14-day usage threshold for refunds?

No. The Refund Policy provides a first-paid-purchase guarantee during a 14-day evaluation window, with reasonable one-time and abuse protections.