Been building some knowledge bases for myself. Tried out several tools like Notebook LM, Afforai and TextCortex. I tested them against each other and Socrates with their deep dive functionality has the most thorough retrieval.

This was my test: I uploaded a markdown document with 55 articles with more than 117K words. Each article title preceded by a # sign. Then asked the tools to give me a list of the articles.

The results were:
Afforai returned 12 names
Notebook LM returned 26 names.
Socrates returned 50 names

I checked the source file and there were some some article names scraped like "\#\[Algorithm Analysis" instead of "#Algorithm Analysis". Probably that caused Socrates to return 'only' 50 of the 55 article names. Anyway 50 of 55 is still impressive and sets Socrates apart from other ai tools.

My review:
Positives:
+ Deep dives are really deep. Deeper than others.
+ I found the developer to be actively responding to queries.
+ Prompt loop where you can ask a promp on each paragraph, section or chapter is really a nice function
+ Table AI where you can compare documents in a table is usefull

Negatives:
- Limited use of other LLM
- UI one has to get used too
- Naming of files. If you upload a file by url it is called file.pdf. To be able to rename it would help.

I'm really looking forward to BYOK preferably with OpenRouter. This would enable the use of other LLM like Gemini or Claude. I found those models a bit more "insightful".

Also looking forward to url scraping functionality. Would be nice if it would enable a bulk url upload. And that you can rename the source.

Maybe it would be helpfull if one could assign an allready uploaded and deepdived document to another chat? In other tools I noticed that you could create "knowledge banks" and that you could chat with one or more "knowledge banks".