Automate Accounting Workpapers with Ledger PDF

Okay, today I want to talk about another app that I built.

This one's called Ledger PDF.

It's meant to be a workpaper binder built for tax and accounting professionals to use alongside AI.

So I think we're all very familiar with creating our work papers in either Adobe or some other application.

Getting the work paper like we want it is huge time suck, right?

It takes, takes a lot of time, takes a lot of effort, a lot of finagling, to get it just like we want it.

And so I create a Ledger PDF to be able to work alongside a workbook with AI and using agents.

So if you saw my video or my other app, Ledger tb, this is a similar theme.

And is also meant to work well with Ledger TV in the sense that once you do someone's books, Ledger tb, you can export out the financials and build out your work paper, from those financials, and have that as your support.

And so this is what Ledger PDF is meant to accomplish.

And you can also use it for other engagements as well.

So again it's meant to work alongside of AI and your agent of choice.

But it doesn't have any AI built into the app itself.

How it works is through mcp, so you can connect your agent via MCP and you can have it work inside the binder and use the tools and all the other features of the binder to kind of build out your work paper.

And because I wanted AI, to be able to help me build out my work papers, I also wanted to be able to clearly see what the AI has done inside the work paper.

And so I wanted to be able to come into a work paper that AI has built and be able to see all the actions that it took and the audit trail basically of what it did, did to create this, this binder.

And so that's what I built in.

You can kind of see a screenshot here, here's a review center.

So you can see basically what the AI did, what it created inside this binder.

so here's some more screenshots of the app.

this is all a fictitious, made up, clients and data and information, completely fictitious.

But you can see similar style to a normal work paper that we build out.

in this profession you got your bookmarks, you have it in order based on you know, whatever the engagement runs by you have tick marks to where you can add calculator, tape, you can add reference numbers, you can add squares, circle, lines, arrows, text.

you can change the status of a page.

You know, all the, the normal tools that we're kind of used to using.

But I think again what's really cool is that you can use AI and an agent to kind of do the first pass of creating this work paper, right?

So you can say Claude, ChatGPT, Codex, whatever you use, go and build out this work paper and you can point it at a folder, with all the documents and you can create your own prompt, your own instructions on how you want your work paper to build.

And the AI can basically use Ledger PDF to build out this work paper and clearly track whatever it's done inside the app.

And this app is again a local desktop app.

You can download it via Windows, you can download it via Mac and you can also go to GitHub and you can review the code yourself.

It's open source, so you can give it to your IT person or give it to your Claude and say, go and check out this repo, see what it is, inspect the code, and kind of decide for yourself whether or not you want to download it and use it.

Okay, so here's how the app opens up.

Just a blank, blank screen.

You can choose which file that you want to start with or you can tell your agent to again compile all the files you need and create the binder itself.

If you see down here, you can see this open binder access on this basically means that the agent can access this binder right now I have it set up via my MCP here with Claude to where the agent can work inside this binder with it being open.

And you can change your agent access down here you can decide which folder you want to give the agent access to.

you can remove access and there's even a command here to where you can connect your own agent and how to do that.

So for example, let's say I wanted to tell Claude again I've made some fictitious client, documents and data.

so I have a couple on my desktop that I'm going to point Claude to and have built out the binder.

So say, okay, let's create the close package for Northline Digital, which again is a fake client using Ledger PDF because I have mcp, connected via Claude.

Claude's going to go and check out what's in this document in this, where it has access to and look for Northline Digital and therefore look what's inside of that folder and take all the documents and compile them into this binder.

So you can see Claude's again reading this document.

It's found the close package and it's reading the financial statements and is compiling everything to start creating the binder.

Okay so you can see here Claude has populated the binder on screen here because I have this open binder access on Claude can work on this screen while I have it open basically so I can see its changes and what it's doing live on the screen.

you know similar to how regular AI computer use does.

But this is done via the MCP and see what I put in here.

It's talked about the engagement, the basis of accounting.

These are some pre filled instructions that I had in the folder that it's working in about what this engagement is, what we're doing for this engagement.

And so it's using all that context to build out this folder.

Some more notes to the financial statements.

It's got my branding for my firm here.

it's got the summary page, income statement page, balance sheet, final trial balance, kind of brought everything in from the month end close package and then it's also brought on the support here.

So if you see these bookmarks over here, you'll see like this is again a fictitious bank activity that Claude built.

But this would be, you know, if you're using this on a real client it would be whatever supporting documents that you're using to do whatever the engagement, whether you're doing a tax engagement or bookkeeping engagement or whatever it may be, it's pulling in the source documents and the final deliverables kind of into one binder and it's going to go through and you can tell it to add things up, add tick marks, circle things square, add tape, and everything will be tracked of what the agent does inside the app.

And so mention about what the AI is doing and being able to see it live.

You can see how right now I'm not could do make any changes into this work paper.

Claude's doing it via the MCP and everything that it does it's being tracked here in this review center.

So you can see it imported this bank of account, the notes, the business checking, it's going through and tying out the cash, tying out the credit card, it's footing the total expenses and everything it's doing, it's tracking inside the app here.

And so you can see here's a calculator tape that Claude did, adding up the total expenses.

19508.

19508.

And it's checking off the net income here.

We'll go back and look at what it's doing here.

If you click down here, you can see again some of the work that it's doing in real time.

Added a Note on page 26, about Adobe Cloud, a note on page 22.

And then it's also will flag open items for review.

So you can see here no depreciation recorded for two months into February 28th.

And so basically I can go to that page and look at note that the agent has left and say, okay, yeah, that does seem off.

I mean it could be fine.

Maybe you don't record depreciation until the end of the year.

Maybe you do it during the months, whatever you may do.

It's basically noting these items inside the workbook and flagging them for your review.

Okay, so if we go back in the app, we can see it's starting to add bookmarks, over here on the right, basically, based on those, all the source documents that it brought in and it's going to continue building out the master file, the master binder.

It's also created, basically a table of contents at the beginning of this binder, showing all of the contents of the binder, what type of file it was PDF, what page it lives on.

And basically you can easily track where every document that was in the folder that you pointed it at ended up inside of the binder.

That way you can easily know, did it bring in all the documents?

Did it miss any, which pages did they go to?

And you can even see that it's tracking some of the marks that it makes on these pages as well next to the document that's listed.

Okay, so it looks like Claude has finished building out, the binder for this CLOJE package.

you can see how it built out all the bookmarks.

You can see all of the items, and actions that it did.

It's got these three open items for review here again pointing out some things that basically just for us to review and go and check, check out.

But I think it's a cool way to a save time trying to build out these master files and kind of offload some of this to AI and then come over top still being the human in the loop and really finalizing this binder and being the final reviewer on record.

And then once you go through it and you mark everything reviewed and you add your own tick marks and your own check marks, you can you know, sign off on the document, you can close it and basically you can, you can export it out as well to where you'll save a final copy of it.

It won't be editable again.

So pretty cool.

again it's open source.

You can download it to your desktop, you can try it out, use it, let me know what's broken, what doesn't work, what could be better.

Would love to keep, keep building this out.

It's pretty rough right now.

I will say it's still some quirks that I'm working through, but I think it's a solid start and could eventually become a really useful tool.