Hey everyone, been testing SuperAGI for a few weeks now. My main goal was to automate scraping and summarizing competitor feature pages for a competitive analysis report.
I set up an agent with a specific goal to visit 10 competitor URLs I provided, extract key feature lists and pricing structures, and compile them into a table. The agent worked, but the results were... spotty.
Out of the 10 targets, it only successfully returned complete and accurate data for 6. The other 4 either had partial info, hallucinated features that weren't there, or failed to navigate the page correctly. A 60% hit rate isn't reliable enough for my use case without heavy manual checking.
I used the default reasoning model and made sure my goals were clear. Has anyone else run into this with web scraping tasks? Wondering if I need a different configuration or if this is a known limitation for now. The potential is exciting, but the inconsistency is a blocker for me.
Learning every day