Troubleshooting

Confirm the script tag is in your page source. Open DevTools → Elements and search for luluclaw.com/widget. If it is missing, the snippet was not saved or deployed yet.

Common causes when the tag is present:

• An ad blocker or browser extension is hiding the widget.
• A Content-Security-Policy header is blocking the script — add https://luluclaw.com to your script-src and connect-src directives.
• The widget only loads on https:// — it will not initialise on plain http://.
• The install token is revoked. Check the Widget tab in your dashboard.

Your widget token likely has an Allowed domains restriction that does not include your production domain. Open the Widget tab in your dashboard, find the Allowed domains list, and add your production domain. Changes take effect immediately.

LuluDesk stores the visitor's session ID in localStorage (key: luludesk_session) for 7 days. History disappears if the visitor clears their storage or uses a private window. This is expected behaviour for anonymous visitors.

The live crawl-progress stream (SSE) verifies workspace ownership. A 403 usually means the session token is stale or you are viewing a workspace you do not own. Try signing out and back in to refresh your session. If you were invited as a collaborator, the live crawl feed requires workspace-owner access.

LuluDesk only indexes content from pages it can fetch publicly. Check that:

• The URL resolves without a login wall or CAPTCHA.
• The server does not return a noindex or X-Robots-Tag: noindex header.
• JavaScript-rendered content: the crawler does not execute JS. If your content only appears after JS runs, use the WordPress integration or the Insights API to push content directly.

Large sites can take several minutes. Refresh the Sources tab after a few minutes. If the status does not change after 10 minutes, delete the source and re-add it. If the issue persists, contact support via the in-app feedback button.

The AI Audit passes your indexed content through a large-language-model analysis window. If the audit output is short or logs a token-limit error, the total content may have exceeded the analysis cap.

A platform update in May 2026 raised the token cap to 4,000 and added better error logging. Click Re-run Audit in the Audit tab to get a fresh report. If it still truncates, try reducing the number of active knowledge sources.

Check your knowledge sources in the Sources tab. A source must show Ready before its content is used. Sources in Pending or Indexing state are not yet searchable.

If sources are Ready, verify that your question is close to the content in your knowledge base. The assistant uses semantic similarity to retrieve relevant chunks — a question phrased very differently from your source content may not retrieve good matches.

This is usually a soul configuration issue. Open the Soul tab and add explicit limits to your soul document. The more specific the instruction, the more reliably the model follows it. Example: "Never quote specific prices. If asked about pricing, direct the visitor to our pricing page."

Add a faithfulness rule to your soul. Example:

"Answer only from the indexed knowledge base. If you do not have a confident answer, say 'I'm not sure — let me get a human to help' rather than guessing."

See the Souls docs for authoring guidance.

The most common cause is a credentials mismatch. In your WordPress admin, go to Settings → LuluDesk → Knowledge Base and confirm the API Key and Webhook URL exactly match the values shown in your LuluDesk dashboard. Disconnect and reconnect the site to generate fresh credentials if needed.

Check the webhook log in Settings → LuluDesk → Webhook log. A 200 response means LuluDesk received the update. A 401 means the API key is invalid — regenerate credentials. A network error or timeout usually means a firewall or WAF is blocking outbound requests from WordPress to luluclaw.com.