skip to content
Site header image Nerdy Momo Cat

I tried Stackie

Stackie is a prompt your own database to obtain prompted structured extraction app, and I am here for it

Last Updated:
Tools AI Beta Testing

As of Nov 5, 2024, Stackie has multiple Twitter posts that could be interpreted as endorsements of Trump. I want to clarify that my testing of this app occurred before I was aware of that position, and it doesn't reflect the values I hold dear.

Stackie (FAQ) is a new app that was suggested to me on Twitter feed from HeyDola (another app that I use for scheduling from images and text, because it works with whatsapp and is free). And it looked fun! It is very similar in idea to what I have implemented in Use Notion’s Property Description As Text → DB add-itor (the example there being a money tracker) and comes halfway through what I mentioned in The Components of a PKMS. But this is a self contained app, has way better UX than hosting your own script, is clean and somehow really clicked for me, because it comes really close to what I wanted (want?) to make in Trying to Build a Micro Journalling App.

To be honest, Notion's AI properties and Notion’s AI add option will get you there pretty often. It is probably too much for you would want if all you are looking for is tracking. There have been other apps that do something similar — hints being the one I can recall off the top of my head, but they all integrate with external apps or are meant for power users or developers (for example, AI add to Supabase database).

When you open the app it starts with a baseline of inbox database. It comes with its own templates, and ideally you should be prompted to select at least one during onboarding to get a feel of how it works. The templates are prompted databases, where each field can either be a date/time, number, boolean or text. The templated database and properties are all customizable which is a huge win!

The entry box when you have created all your “stacks” let's you type in anything and chooses which stack it is most likely to belong to — another affordance I really appreciate. It works with photos too, both understanding the text in the photo (so you can capture a snippet of an event you attended if you are tracking all events you attend in a month), and understands the objects in the photo — so you can click a photo of a cheeseburger and it will understand that it should go to the calorie tracking stack and figuring out the breakdown of nutrients for that log. And it works with voice, so you can speak and it will transcribe and process that information. It seems to use internal dictate option, so doesn't seem to be as good as whisper (proper nouns are hard for example) — but I might be wrong about their processing mechanism.

It can process into multiple databases and add multiple entries at once! It seems to only be additive at the moment though, you cannot edit entries through the universal text box (you can go to the entry and edit it though). There is no export option, but that disappointingly seems to be the norm for iOS and beta apps. You currently cannot do anything with the data you record like you can do in Notion (add it up, set limits etc), so it might not be satisfying to use as a habit tracker and hard to get a view of data you might want, but it is a great starting point. It is what Collections DB could look like, with integrated AI. The app is iOS only, so wouldn’t be something I use, but definitely something worth looking at.

Some images from the app
This screen displays all my current stacks: Inbox, Calorie Record, and Mood Tracker. Inbox is a special pre-existing stack where you can create notes without assigning them to a specific stack. Later, you can combine these notes into a new stack if desired. Calorie Record has multiple properties, such as food input, unit, meal time, and more. The Mood Tracker simply contains two values: the recorded time and the general mood.
This screen displays all my current stacks: Inbox, Calorie Record, and Mood Tracker. Inbox is a special pre-existing stack where you can create notes without assigning them to a specific stack. Later, you can combine these notes into a new stack if desired. Calorie Record has multiple properties, such as food input, unit, meal time, and more. The Mood Tracker simply contains two values: the recorded time and the general mood.
You can create a stack through prompting, which is similar to building a database but using natural language instead. This approach allows you to specify properties, and conversions into structured formats. It's an excellent method for creating a database without using the detailed user interfaces found in tools like Collections or Notion.
You can create a stack through prompting, which is similar to building a database but using natural language instead. This approach allows you to specify properties, and conversions into structured formats. It's an excellent method for creating a database without using the detailed user interfaces found in tools like Collections or Notion.
The best part is the option to create stacks from templates. These templates come with pre-prompted fields and extraction prompts, ranging from AI-generated content for all fields to logging and note-keeping with multiple date properties.
The best part is the option to create stacks from templates. These templates come with pre-prompted fields and extraction prompts, ranging from AI-generated content for all fields to logging and note-keeping with multiple date properties.
So, for example, if you select the calories record, you will see the stack name, the prompt stack indicating which properties should be there, and each property by itself. You can have any number of properties, and you can edit and rearrange them as needed.
So, for example, if you select the calories record, you will see the stack name, the prompt stack indicating which properties should be there, and each property by itself. You can have any number of properties, and you can edit and rearrange them as needed.
Here, I click into a property in the calorie record named "calories." If it's not provided in the input text, the prompt estimates its value. This property is a number, which I can delete, edit, or change its type.
Here, I click into a property in the calorie record named "calories." If it's not provided in the input text, the prompt estimates its value. This property is a number, which I can delete, edit, or change its type.
Here's what that stack looks like. I've added some random records, not actual data from real life, but this is how it appears. One thing to note is that you can use phrases like "yesterday," "today," or "two days ago," and it will adjust the date and time accordingly. It stores both the specified date and time in the field as well as the creation time, similar to how it would work in a tool like Notion.
Here's what that stack looks like. I've added some random records, not actual data from real life, but this is how it appears. One thing to note is that you can use phrases like "yesterday," "today," or "two days ago," and it will adjust the date and time accordingly. It stores both the specified date and time in the field as well as the creation time, similar to how it would work in a tool like Notion.
I can also prompt or add text that contains multiple instances I want to record in a stack. For example, I'm mentioning that I ate cabbage last night and Maggi yesterday morning. My hope is to record these two instances separately as different food items.
I can also prompt or add text that contains multiple instances I want to record in a stack. For example, I'm mentioning that I ate cabbage last night and Maggi yesterday morning. My hope is to record these two instances separately as different food items.
It does indeed. The app records two separate items at different times, estimating meal times for dinner and morning quite accurately. Overall, it captures the entire arrangement quite well.
It does indeed. The app records two separate items at different times, estimating meal times for dinner and morning quite accurately. Overall, it captures the entire arrangement quite well.
Here's my attempt to try out the Spanish words template. The idea here is that you aren't adding any information at all. You're just adding the word, and all of the other information is actually generated by AI. So it's not just structured extraction; it's generation and extraction combined into one, which is really cool.
Here's my attempt to try out the Spanish words template. The idea here is that you aren't adding any information at all. You're just adding the word, and all of the other information is actually generated by AI. So it's not just structured extraction; it's generation and extraction combined into one, which is really cool.
So, here I added the word "beuno" to the stack. I'm using the general input, so my hope is I would first determine that it needs to be added to the Spanish word stack and then add that information to the stack itself.
So, here I added the word "beuno" to the stack. I'm using the general input, so my hope is I would first determine that it needs to be added to the Spanish word stack and then add that information to the stack itself.
It accomplished that quite successfully, so now I have a Spanish word, created a sample sentence, explored multiple variations, chose the correct approach, and decided on the appropriate action.
It accomplished that quite successfully, so now I have a Spanish word, created a sample sentence, explored multiple variations, chose the correct approach, and decided on the appropriate action.
And this is what my main page looks like now, with a new stack added beyond the calorie and mood tracker.
And this is what my main page looks like now, with a new stack added beyond the calorie and mood tracker.