Export Notebook as Text (in 13.3 &14)

Question

What is the best way to export a notebook as readable plaintext - while preserving as much structure as possible? The motivation here is the need to bulk-ingest notebooks and chunk them (for use in RAG systems).

In such a solution for example:

• textual elements would be stripped of styling
• section headings would be preserved
• cell-tags could be listed underneath cell contents
• For non-text elements - those would be simply replaced with their notebook asset filepaths/hyperlinks
• ...

There were a few legacy answers about exporting to multi-markup/down that are now out of date, but is there any new "best way" to achieve this in v13.3/14?

Answer 1

I recently found myself in a similar situation where I wanted text-only but structured files generated from my notebooks for source control and revision tracking. I could not find an optimal solution, but ultimately I landed on exporting the notebooks as a package (*.m) file.

The following code essentially automates the process of opening each file in the front end, and then selecting "Save As..." in the File menu and choosing the "*.m" package option.

I wrote a helper function to do the following:

• obtain the notebook object for the notebook of interest: first retrieve a list of objects for all currently open notebooks, then open the one of interest (the RunProcess part), then retrieve a list of all open notebooks again, and take the difference;
• generate a new file name with a modified extension, replacing ".nb" with ".m"
• save the notebook in "Package" format with the new name using front end tokens (i.e. menu items);
• close the notebook.

My files were all in a single directory, so I just mapped this function over a list of notebook file names.

dir = SetDirectory["C:\\path\\to\\the\\files"];
files = FileNames["*.nb"];

ClearAll[openSaveNB]
openSaveNB[name_] := Module[
  {nbObject, before, newName},
  before = Notebooks[];
  RunProcess[{"cmd", "/c", name}];
  nbObject = First@ Complement[Notebooks[], before];
  newName = FileNameJoin[{dir, StringReplace[name, ".nb" -> ".m"]}];
  FrontEndTokenExecute[nbObject, "Save", {newName, "Package"}];
  FrontEndTokenExecute[nbObject, "Close"]
]

openSaveNB /@ files

The process itself worked; I am still figuring out if the output is acceptable for my goals, but I thought I'd share the process anyway in case you find it useful.

A couple of caveats: 1) I work in Windows and the code makes the assumption that *.nb files are associated with Mathematica, so that invoking one such file in the shell automatically opens it; 2) in FrontEndTokenExecute[nbObject, "Save", {newName, "Package"}], newName should be a fully qualified file name, with the directory as well, otherwise the resulting file is saved in some default location that ignores the SetDirectory directive.