Troubleshooting
If something doesn’t work as expected, here’s how to fix it quickly.
⚠️ Issue: “My prompt used the wrong model”
Possible Causes:
Prompt classification detected a different intent
Manual override wasn't recognized
Fixes:
Specify clearly: “Use GPT-4o for this.”
OR disable the override: “Use the default model for this.”
⚠️ Research output has no citations
Perplexity may ignore formatting and source citation instructions under model stress.
Fixes:
Add: “Be sure to include sources.”
OR use "research-only" tasks
Check whether model override forced use of GPT instead of Perplexity
(Note: I am aware of issues / quirks with Perplexity's API that frequently prevent the output from clearly listing sources / URLs. I will continue to attempt workarounds and monitor for updates)
⚠️ Audio speaks the system prompt / unwanted text
This usually means the audio step used the full composed prompt.
Fix: Add quotes:
“Read aloud only: ‘This is the text to speak.’”
Or rely on dependent text_generation → audio steps.
⚠️ Image didn’t generate
Common causes:
DALL·E content restriction
Too abstract or unsafe
Fix: Rephrase with safe, allowed content.
⚠️ Slow output
Try shorter prompts or prompts involving fewer steps. Some types of tasks (i.e. audio generation) will naturally take longer than others. While most workflows (approx. >90%) are completed in mere seconds, a few more complicated requests could take up to 5 minutes.
⚠️ Errors in multi-step workflows
Dependencies may be misinterpreted.
Fix: Add explicit sequencing terms:
“First… then…”
“After generating the text…”
🛠 When to Contact Support
If something feels broken or you have suggestions / feedback, please email: 📩 [email protected]
Last updated