Troubleshooting

Most issues have simple solutions. Work through these diagnostics before reaching out to support.

Widget Not Appearing on Your Website

Verify Code Placement

The embed script must be placed just before the closing </body> tag in your HTML. If it's in the <head> or in the middle of the page, it may not initialize correctly.

Check Widget ID

Copy the embed code fresh from your dashboard—don't rely on old saved versions. The widget ID must match exactly. Even a single character difference will cause the widget to fail.

CORS Configuration

The website URL in your widget settings must match the domain where you're embedding. If your site is https://example.com but the widget settings say https://www.example.com, CORS will block the connection.

Browser Diagnostics

Open your browser's developer console (F12), then check the Console tab for JavaScript errors. Look for messages mentioning CORS, widget initialization failures, or network errors. These will point you toward the specific problem.

Ad blockers sometimes interfere with embedded widgets. Test in incognito mode with extensions disabled to rule this out.

Simple Test

Hard refresh your browser (Ctrl+Shift+R on Windows, Cmd+Shift+R on Mac) to clear any cached versions of the widget script. This fixes surprisingly many issues.

Bot Not Responding to Messages

Knowledge Base Check

Lumi operates in strict RAG mode. If your knowledge base is empty or doesn't contain information relevant to the user's question, the bot will say "I don't have that information." Add at least 5-10 comprehensive memories covering the topics users are likely to ask about.

RAG System Status

Navigate to the Gitbook tab and click "Test Google AI". This verifies that embeddings are working. If this test fails, RAG is broken and the bot cannot search your knowledge base. Contact support if you see errors here.

Message Quota

Check your usage on the Dashboard or Billing page. If you've hit 100% of your monthly quota, the bot will stop responding. Upgrade your plan or wait for the next billing cycle.

System Prompt

Make sure you've configured a system prompt. While the bot can technically function without one, it won't have clear instructions on tone, style, or behavior.

LLM Model Selection

Verify that a model is selected in your widget settings. The default is llama-3.3-70b-versatile, which works well for most use cases.

Network Connectivity

Check your internet connection and try sending a message from a different network. Browser console errors mentioning network failures or timeouts indicate connectivity issues.

Bot Always Says "I Don't Know"

This is the strict RAG mode working as designed—the bot refuses to answer when it can't find relevant information in your knowledge base.

Improve Memory Coverage

Add more memories that directly address the questions users are asking. Be specific and comprehensive. A memory saying "We're great!" won't match "What are your business hours?"

Test Similarity Search

Think about how the RAG system processes questions. It converts questions to embeddings and finds the most similar memory. If your memory says "Operating hours: Monday-Friday 9-5" but the user asks "When are you open?", the similarity might be below threshold.

Add variations:

Adjust Link Confidence Threshold

In Gitbook RAG settings, lower the link confidence threshold from 50% to 30%. This makes the system more liberal in matching memories to questions.

Check Memory Quality

Short, vague memories don't match well. Each memory should be 100-500 words, focused on one topic, written in natural language.

Voice Messages Not Working

HTTPS Requirement

Voice recording only works over HTTPS. If your site uses HTTP, browsers block microphone access for security reasons. There's no workaround—you must use HTTPS.

Browser Support

Chrome, Safari, and Edge support voice recording. Firefox has limited support and may not work reliably. Internet Explorer is not supported.

Microphone Permission

Look for a microphone icon in your browser's address bar. Click it and ensure the site has permission to access your microphone. If you denied permission earlier, you'll need to grant it again.

After granting permission, refresh the page and try recording again.

Transcription Failures

Speak clearly and minimize background noise. Very short recordings (under 1 second) may fail. Keep messages under 30 seconds for best results.

Audio files are automatically deleted after 3 days. If you're trying to play back a very old voice message, it may have been purged.

LINE Integration Issues

Webhook Verification

Go to the LINE Developers console, find your channel, and navigate to Messaging API settings. Look for the Webhook URL field and click Verify. You should see a success message. If verification fails, double-check that you copied the exact webhook URL from Lumi.

The webhook URL format should be: https://[project].supabase.co/functions/v1/make-server-36a84ccb/webhooks/line/[integration-id]

Enable Webhooks, Disable Auto-Reply

In LINE console, ensure "Use webhook" is toggled ON. Then scroll down and toggle OFF both "Auto-reply messages" and "Greeting messages". These LINE features conflict with Lumi's AI responses.

Token Issues

If messages suddenly stop working, your channel access token may have expired. Generate a new one in the LINE console and update it in Lumi's integration settings. Don't forget to save.

User Profile Not Showing

Check that your LINE channel has permission to access user profiles. This is in the channel settings under Permissions. If disabled, Lumi can still process messages but won't have the user's name or avatar.

Facebook Integration Issues

Webhook Callback Verification

In the Facebook Developers console, go to Webhooks and add your callback URL. You'll need both the webhook URL and verify token from Lumi. Copy them exactly—even extra spaces will cause verification to fail.

Subscription Fields

Make sure you've subscribed to messages and messaging_postbacks webhook fields. Without these, Facebook won't forward messages to Lumi.

Page Subscription

Your Facebook page must be subscribed to your app. In the app dashboard, find your page in the Messenger settings and ensure it's connected and subscribed.

Testing

Send a test message to your Facebook page from a personal account (not the page owner account, as those sometimes behave differently). The bot should respond within a few seconds.

Rate Limiting

If you see "Too many requests" or "Rate limit exceeded" errors, you're hitting API limits. This is rare because Groq API keys auto-rotate across four keys, but it can happen during traffic spikes.

Wait 60 seconds before trying again. If rate limiting persists, you may be experiencing a genuine API outage—check Groq's status page or contact support.

Performance Issues

Slow Response Times

Response time is typically 1-3 seconds with RAG, 0.5-2 seconds without. If you're seeing 10+ second delays consistently, check your internet connection first.

Large knowledge bases (over 1,000 memories) can slow down RAG search. Consider consolidating or removing duplicate memories.

Extremely long system prompts (over 2,000 characters) increase processing time. Keep prompts focused and concise.

Slow Widget Loading

The widget should load in under 1 second. If it's taking longer, check that the embed script is placed just before </body> and has the async attribute.

Look in the browser console for errors or warnings that might indicate network issues.

Payment Failures

Card Declined

Verify your card has sufficient funds and hasn't expired. Check that the billing address matches your card's registered address. Some banks block international transactions by default—contact your bank to authorize charges from Stripe.

Updating Payment Method

Go to the Customer Portal from the Billing page, add a new card, set it as default, and remove the old card. Try the payment again.

Persistent Issues

If your card continues to be declined despite having funds, try a different card. Some prepaid cards and virtual cards don't work with subscription billing.

Browser Compatibility

Supported Browsers

Chrome 90 and newer, Edge 90 and newer, Firefox 88 and newer, Safari 14 and newer. Older versions may work but aren't officially supported.

Not Supported

Internet Explorer is not supported in any version. Users on IE will see errors or broken layouts.

Voice Features

Voice recording works in Chrome, Safari, and Edge. Firefox has limited support. Voice playback works in all supported browsers.

Gitbook Import Stuck

If a Gitbook source shows "Processing" for more than 10 minutes, something likely went wrong.

Auto-Refresh

The interface auto-refreshes every 5 seconds while imports are processing. Make sure you're seeing these refreshes—if the timestamp isn't updating, try manually refreshing your browser.

Force Complete

Click the Force Complete button (checkmark icon) on the stuck source. This marks it as completed without corruption. The system will have imported whatever it successfully processed before getting stuck.

Delete and Retry

If forcing completion doesn't help, delete the source and re-import. Make sure the Gitbook URL is publicly accessible—private Gitbooks can't be scraped.

Check Console

Open browser developer tools and look in the console for error messages during import. These will help support diagnose the issue if you need to contact us.

When to Contact Support

If you've worked through the relevant sections above and still can't resolve the issue, we're here to help.

Before Contacting Support

Gather this information to speed up resolution:

  • Clear description of the problem and what you expected to happen

  • Steps to reproduce the issue

  • Screenshots showing the problem

  • Browser name and version

  • Operating system

  • Widget ID (if relevant)

  • Any error messages from the browser console (F12 → Console tab)

Contact Information

Email: [email protected]

Response Time

Free and Starter plans: 24-48 hours Pro plan: 12-24 hours Enterprise plan: Under 4 hours (under 1 hour for urgent issues)

Business Hours

Monday through Friday, 9am to 6pm Indochina Time (ICT)

Last updated