deep-research

As of now—literally today they announced—OpenAI gives 10 queries of Deep Research per month for Plus users. I wanted to create an alternative, so I took inspiration from this repository https://github.com/dzhng/deep-research but didn’t want to pay for the Firecrawl API too. Instead, I used algorithm from that repository. Here’s the result: https://github.com/jolovicdev/shandu. It’s still rough around the edges—things like proxy support for rate limits aren’t fully implemented yet—but it’s functional! You can try it out with a model you’re running in LM Studio. I’m not entirely sure if this is the best way to build something like this (arch wise, especially since I haven’t worked with LangChain or LangGraph), but here it is - built with a little help from Grok and DeepSeek! Compared to dzhng/deep-research, which relies on the Firecrawl API

All - I'm super excited about this feature! It's an attempt to actually mimic deep research. My repo Open Deep Research has been getting some traction riding on the coat-tails of OpenAI's marketing. :D As flattered as I am about my repo getting some attention, I feel the way I initially set it up wasn’t really deep research. It was shallow research—aka, you have one forward pass: you search for a query, you scrape, and you synthesize (SSS—that's my marketing term for it). But in reality, you SSS, then you have follow-up questions, and sometimes you go down rabbit holes. I was really inspired by this other repo: https://github.com/dzhng/deep-research So, I wanted to see if there’s a UI that can capture this workflow, and I landed on flowcharts. The idea is that a user can come in, do SSS (search, scrape, and synthesize a report for a query