[VIDEO TRANSCRIPT: So this is a fully local web research assistant inspired by Google's deep research. So what you're going to do is go to this configuration. You can put in any local LLM hosted by a LLAMA. It can set how many times you want it to loop. Put in any research topic. Just hit submit. Now it's going to use whatever local LLM I select. In this case, DeepSeek AB is going to generate a search query. It's then going to run web search. It's going to summarize the results from web search. And this is where you can see R1 thinking. And you can see the results stream. Now it reflects on what it learned to generate a new question. Kicks off another round of web search. Gets new documents. Thinks about them to extend the summary. Build more knowledge. Now what's cool about using R1 is you can really see its thinking pattern. Updates the summary and its learnings. Reflects again. Identifies knowledge gaps. Generates a follow-up question. Does more web research. And it'll do this loop as many times as you can figure. Now this is all running locally. So there's no API costs associated with it. It'll just burrow as long as I want. So it's a very fun way to kind of see how R1 is thinking through problems. And I found that the 8-billion model runs pretty well. I'm on a 32-gig MacBook Pro with M2 Macs. And you can see it really builds a quite nice summary. In this case, I'm asking about a particular algorithm, PPO. And it finishes. So when it finishes, it adds a little bit of formatting. And it adds all the sources that it gathered from web research. The whole thing ran about a minute. And it can use any local model configured through Olama that you download. Just set the configuration. Pick whatever model you want. I found R1 8-b is quite nice. You can see it produces really good summaries. And what's nice is you can watch this entire reasoning process going through this iterative iterative research summarization identify knowledge gap loop.] Someone built a free local alternative to Perplexity Pro Writes its own search queries, finds its own gaps, keeps looping Outputs a full markdown report with citations Works with DeepSeek, Llama, Qwen via Ollama 8,500 devs already starred it. Costs $0 https://t.co/pEqZ3sXwPn