Performance and limits
How a workflow performs depends mostly on where it runs. Local runs happen inside a single WordPress web request and are bound by your host’s PHP limits. Cloud runs happen on our engine and are not.
PHP execution limits (Local mode)
In Local mode, a workflow runs inside PHP, so it is subject to your host’s maximum execution time and memory limit. Long AI calls, large documents, or many steps can hit those limits and end a run early, especially on shared hosting.
If you regularly hit timeouts, switch that workflow to Cloud. Cloud runs have no PHP timeout and are scheduled server-side. See Local vs Cloud execution and Troubleshooting.
Scheduling reliability
Scheduled Local runs depend on wp-cron, which only fires when your site gets traffic. Low-traffic sites can see late or missed runs. Cloud scheduling fires on time regardless of traffic. The troubleshooting guide covers both fixes.
Loop safety cap
The Loop node repeats steps over a list or until a condition is met. It has a built-in safe cap on iterations so a runaway loop cannot lock up a run. Keep loops bounded and prefer a clear stop condition.
Credit costs per node (Cloud)
Local runs with your own key are free. Cloud runs are metered in credits per action. Simple structural nodes cost nothing; heavier AI and media nodes cost more. Current costs:
| Node | Credits per run |
|---|---|
| Trigger, Output, Condition, Loop, Parser, Create Post, Human Input | 0 |
| API Call, Send Email, Connect an App | 1 |
| AI Sentiment Analysis | 1 |
| AI Prompt (AI Model), AI Summary, AI Extract Information, AI Optimize for SEO | 2 |
| Create File, Generate PDF | 1 to 2 |
| Write an Article | 3 |
| Chatbot (per agent turn) | 3 |
| Unsplash Image | 3 |
| Web Scraper (Firecrawl) | 5 |
| Media Generator | 10 |
| AI Research (Perplexity Sonar) | 40 |
The Research and Media Generator nodes are the most expensive, so budget for them. Watch your consumption in the Usage Center. See Cost and usage tracking and Your account and credits.
Tips for faster, cheaper runs
- Keep prompts tight and give the AI only the input it needs.
- Use the AI node’s structured multi-output instead of adding extra Parser or Extract steps.
- The workflow generator and in-canvas assistant use prompt caching, so repeat calls are much cheaper.
- Run heavy or scheduled workflows in Cloud, and keep quick, private ones Local.