langchain-ai/local-deep-researcher — reverse-engineered prompt

Build me a fully local research assistant that can take any topic I type in, search the web, read the results, summarize what it found, notice what is still missing, then run a few more searches to fill the gaps. At the end it should give me a clean markdown report with the sources it used.

I want it to work with local models through Ollama or LMStudio, so I can choose the model in an environment file. For search, start with DuckDuckGo because it should work without an account, but make it easy to switch to Tavily, SearXNG, or Perplexity if I add keys. Let me set how many research rounds it runs and whether it fetches full page content.

Please include setup instructions for Mac and Windows, a sample environment file, and a simple way to run it locally through LangGraph Studio so I can watch the research steps happen. Look up current docs online if you need to.