Perplexity
Using Perplexity Sonar models in Aimogen Pro for web-grounded generation, when to prefer them over the plugin internet access feature, and their limitations.
Perplexity Sonar models search the web as part of answering. That makes them useful when factual currency matters more than stylistic control.
Get a key#
- Sign in at perplexity.ai and open API settings.
- Generate a key and add credit.
Configure#
Aimogen Pro › Settings › API Keys › Perplexity AI API Keys
One key per line, Test API key, save.
Models#
| Model | Character |
|---|---|
sonar | Fast, light web-grounded answers |
sonar-pro | Higher quality, more sources |
sonar-reasoning | Adds a reasoning pass |
sonar-reasoning-pro | Reasoning plus the stronger retrieval stack |
sonar-deep-research | Long-running, multi-source research. Slow and comparatively expensive |
When to use Perplexity#
Good for
- News summaries, "current state of X" articles, anything where stale training data is a real problem
- Research passes that feed a second model for final drafting
- Fact-oriented sections inside a larger article
Less good for
- Long-form creative writing. Sonar output is answer-shaped, not article-shaped
- Precise formatting control. Structural instructions are followed less strictly than by general models
- High-volume bulk generation. Grounded calls are slower and pricier per post
Perplexity versus the plugin internet access#
Aimogen Pro has its own AI Internet Access feature: it runs a search, scrapes the results and injects them into the prompt for whichever model you chose.
| Perplexity Sonar | Plugin internet access | |
|---|---|---|
| Who searches | Perplexity | Your server |
| Works with any model | no | yes |
| Control over sources | low | high — you choose the backend and the result count |
| Extra credentials | Perplexity key | Optional SERP key, or free scraping |
| Latency | one call | search plus scrape plus generation |
Using both at once is usually wasteful. Pick one.
Limitations in the plugin#
- No embeddings. Use OpenAI, Google or Ollama.
- No image generation.
- Tool calling is not part of the Sonar design, so Perplexity is a poor fit for Chatbot Extensions and agents.
sonar-deep-researchcan exceed PHP execution limits. Prefer it in a background context.
Common problems#
Timeouts on sonar-deep-research
Expected. Raise PHP max_execution_time, or use a lighter Sonar model.
Output includes citation markers That is the model style. Strip them with a post-processing regex in the rule, or instruct the model not to cite inline.
401 or 429 Wrong key, or tier limits. See API and provider errors.
Endpoint used#
https://api.perplexity.ai/chat/completionsRelated#
Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.