In this video I test OpenAI’s new Responses API inside n8n, and I genuinely think we’re about to see a big decline in people relying on tools like Perplexity or Tavily just to power agent “research.” The big issue has always been freshness—LLMs lagging 6–12+ months—so I walked through the web search tool first and stress-tested it with stuff that literally happened hours ago. I also show the practical controls you actually care about in workflows: search context size, narrowing by city/country for localized research, and even restricting results to a specific domain (like ESPN.com) when you want clean sourcing. Then I move into file search, which is basically a dead-simple RAG setup without you building a whole vector DB pipeline. I create a vector store in the OpenAI API, attach a small ultrarunning guide, drop the vector store ID into n8n, and query it for a 100-mile training question. It works fast, but I also call out the cost model (like the $0.10/day once you’re over 1GB) and why I’m still curious how this compares to Gemini’s cheaper file search or a “normal” RAG pipeline. Finally, I demo code interpreter: turn it on, send a cleanup task, and it writes/runs Python to clean a CSV and produce a downloadable result—perfect for data-heavy automations.