I built a rules + AI categorization app for Tiller that runs locally on your Mac — would this be useful?

Conversation Ask Anything
1

Hey all — longtime Tiller user, first time posting a “I built a categorzing thing.” I’d genuinely like to know whether this solves a real problem for others or just for me.

The itch. I wanted categorization that could be as simple or as complex as I needed, that I could edit anytime, that is transparent+obvious about what is happening and that could get AI to help write the rules in the first place. I tried using others’ and rolling my own with Google Apps Script and hit three walls: it didn’t carry over when I started a new spreadsheet, it was very slow to run, and it was painful to debug. So I built a desktop app instead.

(I could only have one screenshot so I combined them)

categorify-all-screenshots
categorify-all-screenshots1920×2025 359 KB

What it is. Categorify is a native Mac app (soon windows also) that runs against your existing Tiller Google Sheet (soon also Excel). It’s rules-first: a deterministic engine categorizes everything it can, instantly and locally. Only the genuinely new transactions get sent to AI (Claude), which proposes a category and drafts a reusable rule with a confidence score. You approve with one click — and that merchant is handled automatically from then on. The rules get better, and the AI gets cheaper, every month.

A few things I care about:

  • Rules you can actually read and edit — conditions on merchant, description, amount, account, or recurring cadence. The AI can write the regex so you don’t have to.
  • Fast and local — it mirrors your sheet’s transactions into a local database, so it’s instant and works offline. No waiting on Sheets, nothing to debug.
  • Non-destructive — Tiller still owns your categories. Categorify writes into them; it never rewrites your taxonomy.
  • Portable — your rules aren’t trapped in one spreadsheet. They come with you to the next sheet or client.
  • Private — your data stays on your Mac/local computer; API keys live in the OS keychain.
  • More private - data never leaves your computer beyond the app working directly with the spreadsheet. No 3rd parties are involved.
  • use your own ai agent+llm for creating categorization rules , could even run your own local LLMs.
  • Amazon order history categorization

I read the community AutoCat threads. A lot of what I saw is on my roadmap: starter rule packs so you’re not staring at a blank sheet, automatic transfer matching (paired debit/credit → Transfer), amount-aware rules (Amazon $12 vs $220), splitting one charge across categories, rules that set the category and clean up the description, and bulk “rules from your last 90 days.” If it’s realistic, I want to build it — full list here: (can’t link, but if you want to see the site categorify-site (dot) vercel (dot) app

What I’m asking:

  1. Would a tool like this actually save you time, or do you prefer to keep everything in the sheet?
  2. What’s your single biggest categorization headache today?
  3. Which roadmap item would you want first?

It’s early and I’m still developing the Mac-only right app now — I will open up early access if there is interest and would love feedback from folks who categorize a lot.

I did see the latest beta updates for categorization in the screencast. I think those will be great for a lot of people, even most. But I think a greater level of transparency, speed and predictability will appeal to some users. If valuable I’d also be delighted to work with Tiller directly to make this accessible to everyone.

Thanks for reading — honestly, this community’s AutoCat threads basically wrote my to-do list. :folded_hands:

2

Ultimately one more difference between this and other DIY solutions posted is that it will be much easier to set up and require no ‘programmer’ skills per se, though Google honestly does not make that easy!

3

Very ambitious! Sounds like an interesting tool. I personally like to touch every transaction by adding the category manually except for the most basic stuff that I create rules for in AutoCat. Judging by the amount of categorization posts, it’s more important to many others. I wonder though how many people are ready to jump into AI with this. It’s still very early in the game, so your uptake may be slow, but if what you’re offering works better than what’s being built into the Tiller workflow, you may have some takers. Good luck!

4

@brendan2 Great project. I think the database native to Macs is MySQL. Is that what you are using? Unfortunately, I was limited as to what I could see in your screenshot. (It didn’t translate well in IOS) but will check it out later on my M4 MacBook Air (unfortunately my cat ate through the usb-c to MagSafe power cable) when I get it up and running on Thursday. It’s real hot here or I’d use my Windows 11 laptop to see it. Anywho, I had mixed results with Tiller’s AI Suggest and Description Match when I had my Google Sheet running. (On an excel template now). Is there an error rate for what AI Claude suggests? Is Gemini Pro an Option? What I did in my own SQL database was to utilize Category Hints (from the aggregators) where there was no category selected on the transaction sheet. May be that your approach is better. I’m interested in trying it out if I could utilize Gemini pro and I could help test on my Mac if you are looking for volunteers. -David

5

Great project.

Thanks!

I think the database native to Macs is MySQL. Is that what you are using?

Many different databases run on MacOS, basically all databases. And I’m using SQLite, but may consider evolving to Turso at some point.

Unfortunately, I was limited as to what I could see in your screenshot.

if you go the site i mentioned in the post (couldn’t post a clickable link) you can see the screenshots.

(It didn’t translate well in IOS) but will check it out later on my M4 MacBook Air (unfortunately my cat ate through the usb-c to MagSafe power cable) when I get it up and running on Thursday. It’s real hot here or I’d use my Windows 11 laptop to see it. Anywho, I had mixed results with Tiller’s AI Suggest and Description Match when I had my Google Sheet running. (On an excel template now). Is there an error rate for what AI Claude suggests? Is Gemini Pro an Option? What I did in my own SQL database was to utilize Category Hints (from the aggregators) where there was no category selected on the transaction sheet. May be that your approach is better. I’m interested in trying it out if I could utilize Gemini pro and I could help test on my Mac if you are looking for volunteers

Sounds great. I will be. First wanted to gauge interest. =)) This is clearly a big need in the community given all the solutions people have created, though mostly programmers for programmers. This would ultimately be installable and usable like any other desktop app and ultimately for both Mac and Windows if people are interested. Thank you!

6

Very ambitious! Sounds like an interesting tool.

Thanks! It was very ambitious pre-vibe coding. I’m a software developer but have recently created a few apps in a few days that would have taken months or a year. Recently created a niche one called “beadmeupscotty” which you can google (free, open source, niche for vibe coding).

I personally like to touch every transaction by adding the category manually except for the most basic stuff that I create rules for in AutoCat.

I think everyone’s use case is slightly different. I have thousands of transactions per year, no one would want to touch them all. Second marriage, contractor across multiple accounts, etc. But I’m looking to make something easy to use but powerful…like Tiller is, but just very specific categorization tool.

Judging by the amount of categorization posts, it’s more important to many others. I wonder though how many people are ready to jump into AI with this.

This is a good question. I think the answer is most people. And in reality people won’t be passing that much data to ai, just transaction descriptions, amounts, maybe accounty type, could be configurable. And with this tool once the matching rules are in place ai is not touched. Soon local ai will be powerful enough.

From what I’ve seen and how willing people are to let existing cloud services access all their data I think few people care or are aware or either or both.

This is not really an ai tool, its a tool that will use ai to create rules one time, probably a few dozen rules on average. TBD.

It’s still very early in the game, so your uptake may be slow, but if what you’re offering works better than what’s being built into the Tiller workflow, you may have some takers. Good luck!

[/quote]

Thanks!

7

@brendan2 i tried on iOS in safari to go to the site categorify-site.vercel.app and I got a warning. I did not proceed.

8

Thanks for letting me know. I’m not sure why. I tried it on my iphone off your post in safari on iphone and got no warning. ¯_(ツ)_/¯

I think it might have to do with the email field. i’ll look into it.

9

Just tried it again. It’s fine. Must have been a typo! -David