LiquidText and OpenAI Integration for PDF Files

I am a long-term LiquidText user for annotating PDF files. Note you can save LiquidText Files in DT3 and “Open With” to open them in LiquidText

This announcement is a very nice demo of something we have discussed… searching long PDFs and getting links back to specific locations.

I need to dust off my iPad to try this out … .the examples look spot-on for the type of use we have been discussing.


The only thing I remember of LiquidText when I tested it some time ago is crashes under crashes, lost annotations, and more crashes. Assuming those issues are solved, take into consideration that if the “AI” used is free ChatGPT 3.x and is not local built-in one, it is extremely limited in PDF size to convert it in a toy for not more than 3- or 4-page documents.

I don’t know for sure but I suspect the Liquidtext crashes are memory-dependent. I use it extensively and have mostly used it on a Mac with immense memory and it works well at least up to PDFs 5,000 pages long.

Their tech support says there is no limit to the number of pages though it analyzes documents in chunks so cannot always get the “big picture.” I suspect that may be fine for my work; I am mostly looking for an advanced search tool rather than a high-level summary.

So using an example I gave in another post, I would find it immensely useful if a prompt like this would work: “An imaging study is an x-ray report, MRI report, CT scan report, nuclear medicine report, or ultrasound report. Show me a list with links of all pages in this document with an imaging study.”

Its ability to give a link back to the actual source paragraph (always a strength of LiquidText) is a huge plus. As is their recent cloud sync which would likely allow bookmarking in DT3 and thus having AI/annotations on the iPad immediately available on the Mac.

I haven’t tried it yet - but will do so soon.

1 Like

It sounds interesting.

Please, put your results here when you try it.

1 Like

I guess in case of long PDF documents (especially up to 5000 pages) such a query would require a huge amount of patience.

That’s a big question indeed I want to explore

But even it it takes a long time, It can operate in the background or can be done on a different device and in that case even if it takes hours it does not matter as long as it works unattended . The resulting LiqudText file can then be saved for use later.

OCR takes quite a bit of time too on such documents. But DT3 has that built-in so it is super-easy to do and thus it does not matter if that takes. hours.

One major difference is that OCR is usually performed once per document but searching/summarizing like that might happen much more frequently.

Agreed - it all depends on the use case. And what is useful for one person will not be useful for everyone.

As for me - while the ability to ask arbitrary questions in real-time would be the ultimate solution, I can create a template of keywords or concepts to search for that apply to the majority of the cases clients send me. Such a list of results - with a link right back to the original source text - would help immensely in my productivity when reading through records and particularly would help as a double-check for me to not miss something important.

Now I suppose I might be able to do something like that writing a script for DT3. I think it is possible but it would be hard for me to come anywhere close to replicating the quality user interface of LiquidText in creating such deep links.

Someday if DT3 gets an “Ai Search” feature like that it would be icing on the cake. But for now it works fine to use DT3 to organize all the documents and search results and as a launch pad for opening LiquidText or other related viewing/editing/indexing apps.

Another concept I have played with a bit is using AI to extract summary information from text not only in paragraph form but instead as JSON or CSV data. That gives the option to then view/search/integrate the data in other ways.

Interesting thread, thank you all.

New user to LiquidText here. I am hoping to learn how better to incorporate LT with DT3 and with Bookends, my bibliography program.

When you note in the original post that DT can open the LT files, does that mean it can read and index them, in your experience? Do you keep the original PDF as well as the LT file that has imported the PDF for annotation.

Also, LT can export annotations in PDF format, and curious if you find this useful in any way in conjunction with DT3. For example, can you index in DT, and then follow a link back to the LT file.

My interest would be to save LT files in either Finder (and index them only) or in DT (if it can index and incorporate within its AI), but allow the LT files to link to other items, like original articles.

Thanks in advance for sharing any of your experience and knowledge!