Sitemap not found, JavaScript-only sites, charts succeeding but returning no content, page count mismatches — common Atlas failure modes and how to fix them.

Common Atlas issues and how to resolve them. If something is off, the source row in Settings → Agent → Abilities → Atlas access shows the most recent error inline (red text under the URL).

"Sitemap not found" (crawl_fallback)

Atlas tried robots.txt, sitemap_index.xml, sitemap.xml, and wp-sitemap.xml, and none responded. It fell back to crawling links from the homepage, which usually means lower coverage (deep pages without homepage links won't be found).

Fix: publish a sitemap. Most static-site generators (Hugo, Astro, Next.js, Docusaurus, Mintlify) generate one automatically; check your build output for sitemap.xml and link it from robots.txt:

User-agent: *
Allow: /
Sitemap: https://docs.acme.com/sitemap.xml

If you can't add a sitemap, you can keep using crawl-fallback mode but accept that only homepage-linked pages will be charted.

Charts succeed, but pages have no content

The page count goes up but the agent never quotes the source, or you open the Pages modal and the rendered Markdown is just nav and footer.

Cause: the site is JavaScript-only. The HTML cStar receives is a shell, and the real content only appears after JS runs.

Fix: edit the source and toggle Render JavaScript on, then Re-chart. JS rendering is slower but works for SPAs. If you're unsure, open one of the chart'd pages, view the source in your browser (Cmd+U / Ctrl+U), and look for the main article content. If it's missing, you need JS rendering.

"Page count mismatches what I see on the site"

A few possibilities:

  • Auto-skip rules: Atlas drops common low-signal patterns by default (WP archives, query-heavy URLs, feeds, admin paths). The Pages modal lists charted pages; check there to see what was kept vs skipped.
  • robots.txt disallows: pages disallowed by the source site's robots.txt are not charted. This is correct behavior and not configurable. Atlas is an honest crawler.
  • noindex meta: pages with a <meta name="robots" content="noindex"> tag are skipped.
  • HTTP errors during chart: the source row shows a count of pages with errors. Open the Pages modal. Pages with a 4xx/5xx will be marked.

"I added a source but the agent never uses it"

Two checks:

  1. Per-agent toggle: a source is team-wide, but each agent has to be opted in. In the agent's Atlas access section, make sure the toggle next to that source is ON for this agent.
  2. The agent has Atlas as a skill: the Atlas access master toggle (above the source list) needs to be on for the agent. If it's off, the agent can't use Atlas at all regardless of which sources are toggled on.

You can also test from the agent's eval tab. Ask a question whose answer is on a charted page and see whether the agent quotes it.

"Re-chart hangs / takes forever"

Re-chart works through pages in batches of 5. A 1,000-page source typically takes 10–20 minutes plain-fetch, longer with JS rendering. A 10,000-page source can take an hour or more. The progress bar updates after every batch.

If a re-chart is stuck (no progress for several minutes), check the source row for an inline error message. Common causes:

  • Source site is rate-limiting Atlas (you'll see HTTP 429 errors). Wait, then re-chart.
  • A network blip; cancel and re-chart.

"I want to remove a source"

Hover the source row in the Atlas section and click the trash icon. Confirms first. Removing a source deletes the chart and removes it from every agent's per-agent toggle list. The external site is not touched.

"I hit the 20-source limit"

Each team can have up to 20 Atlas sources. If you hit the cap and need to add more:

  • Consolidate by giving Atlas a higher root URL that includes multiple sub-sites
  • Remove sources you're no longer using
  • Use exclude patterns within an existing source rather than creating a separate source per sub-section

The 20-source limit exists to keep retrieval latency predictable. We're open to raising it for teams with a real need, so reach out if that's you.

Still stuck?

Open a ticket from the chat widget on this page. Include:

  • The source name and root URL
  • The error message in the source row (if any)
  • Roughly how many pages you'd expect Atlas to find
  • Whether the source site is server-rendered or a SPA