Themeisle Docs

Appearance

Troubleshooting Hyve answer quality ​

If Hyve gives incomplete or irrelevant replies, or does not include useful source links, use this guide to diagnose and fix the most common causes.

Hyve gives incomplete answers ​

Incomplete answers usually happen when your Knowledge Base is not fully processed, the OpenAI connection fails, your content is hard to match semantically, or the selected model is too limited.

1. Confirm your Knowledge Base finished processing ​

When you add or update sources, Hyve needs time to process them into embeddings. This is handled by WordPress cron tasks, which typically run about once per hour.

  1. In your WordPress dashboard, open Hyve > Knowledge Base.
  2. Check whether your recently added content is still pending or listed in Requires Update.
  3. Wait for the next cron cycle, then test the same question again.

📝 Info: If your site has low traffic, WP-Cron may run less frequently, which can delay Knowledge Base updates.

2. Check your OpenAI API key and available credit ​

If OpenAI requests fail, Hyve cannot generate embeddings or reliable responses.

  1. Open Hyve > Settings > Advanced.
  2. Confirm that OpenAI API Key is set correctly.
  3. Verify in your OpenAI account that API usage is enabled and credit is available.
  4. Return to Hyve > Dashboard and review any visible error notices.

3. Use a more capable assistant model ​

Lower-cost or legacy models can return weaker or shorter responses for complex questions.

  1. Open Hyve > Settings > Assistant.
  2. In Model, select a more capable option (for example, GPT-4.1 or GPT-4o).
  3. Save changes and test the same question again to compare answer quality.

💡 Tip: Keep your Temperature and Top P defaults while troubleshooting. Change only one variable at a time so you can clearly see what improves results.

⚠️ Warning: Higher-capability models can have lower tokens-per-minute (TPM) limits on lower OpenAI usage tiers. If the chatbot answers correctly at first but then returns "Sorry" messages after several turns in the same conversation, switch to a model with higher TPM headroom, such as GPT-4o mini. See Hyve shows "Sorry" or stops responding after several messages below.

4. Write Custom Data for semantic matching ​

Hyve retrieves answers by semantic similarity. Your visitor's wording must be close enough to the processed Knowledge Base content, not only to a title.

When you add Custom Data or FAQ-style content, include:

  • the exact question users ask
  • one or more natural-language variants
  • a direct answer in plain language
  • the full URL when you want Hyve to include a documentation link in its reply

Instead of a short generic entry like this:

text
Title: Refunds
Body: We offer refunds.

use a richer entry like this:

text
Title: How can I request a refund?
Body:
How can I request a refund?
Can I get my money back if the plugin did not fit my needs?
To request a refund, follow the steps in this guide:
https://docs.themeisle.com/billing/how-to-request-a-refund

Hyve can only return links that exist in the matched Knowledge Base content. If the source text does not include a URL or clear link target, Hyve may answer without a clickable documentation reference.

To improve link inclusion:

  1. Add the destination URL directly in your source article, Custom Data, or FAQ content.
  2. Re-process any source listed in Requires Update so Hyve stores the latest version.
  3. Ask your test question again and check whether the reply now includes the link.

6. Retest after processing finishes ​

After any Knowledge Base update, run this quick checklist before re-testing:

  1. Open Hyve > Knowledge Base and confirm no relevant items are still pending.
  2. Check Requires Update and process outdated items.
  3. Wait for WP-Cron to complete background processing.
  4. Open Hyve > Dashboard and review any current error notices.
  5. Ask the same question again in the chat and compare the result.

Hyve shows "Sorry" or stops responding after several messages ​

If the chatbot answers correctly at first but returns "Sorry" messages or stops responding after several turns in the same conversation, the most likely cause is an OpenAI API rate limit. Hyve sends the full conversation history with each request, so token consumption grows as the chat gets longer. On lower OpenAI usage tiers, a long conversation can exceed the tokens-per-minute (TPM) limit for some models.

Clearing the chat, refreshing the page, or switching browsers can temporarily restore responses because it resets the conversation history and reduces the token count sent with the next request.

Switch to GPT-4o mini for longer conversations ​

GPT-4o mini supports a higher TPM limit than GPT-4o on Tier 1 OpenAI accounts, making it better suited for sites with longer chat sessions.

  1. Open Hyve > Settings > Assistant.
  2. In Model, select GPT-4o mini.
  3. Save changes and start a new chat session.
  4. Test with a conversation of similar length to the one that triggered the error.

Upgrade your OpenAI usage tier ​

If you need a higher-capability model such as GPT-4o or GPT-4.1, or if your site has high chat volume with long conversation threads, upgrading your OpenAI usage tier raises the TPM limits available to your API key. Review your current tier at platform.openai.com.