Spent a little time deep-diving on deep-research this week. I have been using ๐†๐๐“ ๐š๐ง๐ ๐†๐ž๐ฆ๐ข๐ง๐ข'๐ฌ Deepresearch for most of my research tasks but unfortunately they are not available as APIs. So figuring out which deep-research API or open-source repo on Github is best for our tasks has been a bit of a pain. So I tried to simplify the decision making by evaluating a bunch of them on different tasks and ended up with the following conclusions:

๐๐ž๐ฌ๐ญ ๐Ÿ๐จ๐ซ ๐ฉ๐ž๐ซ๐ฌ๐จ๐ง๐š๐ฅ ๐ฎ๐ฌ๐ž: ๐†๐๐“ ๐ƒ๐ž๐ž๐ฉ๐ซ๐ž๐ฌ๐ž๐š๐ซ๐œ๐ก if you want more ๐ข๐ง๐ญ๐ž๐ฅ๐ฅ๐ข๐ ๐ž๐ง๐œ๐ž, ๐†๐ž๐ฆ๐ข๐ง๐ข if you want more depth in search.

๐๐ž๐ฌ๐ญ closed source API: Perplexity Sonar

๐๐ž๐ฌ๐ญ open source projects:

1. GPT-Researcher by Assaf Elovic
and Tavily (https://lnkd.in/g3V8uPaf) was the easiest to work with, provides decent results but no tool calling and deeper analysis, only uses pure LLM based summarisation
2. Deepresearchagent by Skywork AI (https://lnkd.in/gfN2VHhG?) seems architecturally the most interesting and holistic but difficult to get it working consistently and specifically for your use as it uses so many different components

3. Gemini deepresearch with Langgraph released by Google (https://lnkd.in/g3mNqKhj) is promising and quite easy to start with, provides better results than GPT-Researcher with Tavily at the moment, but again lacks the tool calling which could be necessary for a lot of actual analysis.

But these repos along with Jina's deepsearch (https://lnkd.in/gXM-2XS4) for finding the "right answer" seem most useful to start building with.

The most interesting one I have come across is Exa's websets ( https://websets.exa.ai/), where you can deeper search but get structured output but I haven't been able to use it for professional purposes yet.

Are there any interesting solutions or opensource repos I am missing out on?


This post was originally shared by on Linkedin.