Refleks

Refleks documentation

Troubleshooting without guesswork

Start with the status and diagnostic detail in the dashboard. Then use the matching section below to separate scope rules, a site response problem, a plan cap, and a browser security policy.

A crawl is stuck or failed

Status is still fetching or indexing
  • Fetching reports discovered, processed, and fetched counts. Indexing reports indexed pages and chunks.
  • Large or JavaScript-heavy sites can take longer. The worker also reduces fetch concurrency when recent chat activity or system load requires it.
  • Refresh the site page and open the crawl report for the latest job and last-progress time. A stale running job is eventually marked failed with a timeout message.
The crawl says no indexable pages
  • Open the starting URL in a browser and confirm it is publicly reachable and returns HTML.
  • Check that the page has visible text, not only a client-side shell. A page with fewer than 200 extracted characters is skipped.
  • Check robots.txt, the hostname in your sitemap, and whether important routes sit under an excluded application path.
  • Use the manual text import if the site intentionally blocks crawling or the content is in a supported text file.
Some pages are missing
  • Check the skipped-page list for robots_blocked, unsafe_url, fetch_empty, fetch_failed, content_too_short, alternate_language, canonical_duplicate, or a crawl-budget reason.
  • Confirm the page is on the exact crawl hostname. A subdomain or a redirected alternate hostname is outside the crawl boundary.
  • Confirm the active plan’s pages-per-site limit has not been reached. A sitemap can contain more URLs than the plan permits.
  • If a canonical tag points to another same-host URL, Refleks indexes the canonical representation and skips the duplicate.

The page is rendered by JavaScript

Conditional

Check the deployment’s browser sidecar

Refleks can use Crawl4AI when the deployment enables it, but public accounts cannot turn that service on from the page. If it is unavailable, the crawler falls back to direct HTML and cannot see content that exists only after browser JavaScript runs.

For a reliable fallback, server-render important content, publish it in the initial HTML, add a same-host sitemap entry, or use the dashboard’s manual text path. Do not assume that an interactive browser preview is what the direct crawler receives.

Answers are missing, stale, or inaccurate

  1. Open the indexed pages list and confirm the authoritative page was included in the active snapshot.
  2. Run a refresh after publishing the corrected content. The service hashes extracted content and can reuse unchanged chunks, while changed content is re-indexed.
  3. Ask a question using the same terms as the page. Retrieval is semantic plus lexical; it can miss unusual wording, very short content, or facts outside the indexed context.
  4. Review the supporting URL. Sources are returned when relevant chunks are available, but a source link does not guarantee that every sentence is correct.

If the answer is not in the indexed content, the model is instructed to say it is unsure and direct the visitor to the business. It cannot know an unpublished update or override a stale active snapshot.

The widget does not appear

  • Copy the current script from the dashboard rather than reusing an old key or versioned URL.
  • Confirm the script is published in the global layout and is not restricted to a preview, draft theme, one route, or a consent category that never runs.
  • Run Verify installation. A framework or tag-manager install may be reported as a reference rather than a normal script tag; test the published page in a browser too.
  • Inspect the browser console and network panel for CSP, CORS, blocked mixed-content, visitor-session, or origin-not-allowed errors.

See CSP and origins for the required app-origin allowances and verification details for the probe’s limits.

A request says the limit was reached

Conditional

Site allocation

A downgraded account can keep a stored site outside its current allocation, but crawl, chat, and widget config requests for that site are blocked until the allocation is fixed.
Conditional

Monthly messages

Visitor and dashboard messages share the account-wide UTC-month cap. Dashboard tests also have a 20-per-site/10-minute guard.
Conditional

Daily refreshes

Crawl jobs share a per-account UTC-day cap, and the same site has a five-minute cooldown after a finished or failed job.

A cap response does not delete the active index. Wait for the relevant window, upgrade, remove an earlier site, or contact support with the response code and account email. See Plans and limits for the full table.

When to contact support

Include enough evidence to let us reproduce the boundary you hit.

  • The site URL and the exact page or widget location.
  • The crawl job stage, last-progress time, skipped-page reason, or HTTP status.
  • The expected result and what happened instead, with a screenshot or console message if relevant.
  • Your account email. Never send a password, widget key, visitor token, or API credential.

Email support@refleks.dev. We can help with account access, crawling, answer quality, widget installation, billing, cancellation, and refunds.