Blogs
dig3st-hero.jpg

dig3st : AI Notification Summaries for EVERYONE

u
uncoreMarch 13, 2025
7 minutes to read

You can skip all of this reading by watching this video it would help me a lot thank you

I tried fixing Notification Summaries for EVERYONE (Jailbroken) | dig3st
I tried fixing Notification Summaries for EVERYONE (Jailbroken) | dig3st


1.The inspiration

The inspiration came from noticing how many people struggle with Apple's Intelligence features—check out this Subreddit for more insights. I found myself wondering: is there a way to improve notification summaries ? Apple, being Apple, doesn't allow users to configure the model, system prompt, or user prompt for notification summaries. And while that might seem limiting, there's a reason behind it. Apple aims to keep things user-friendly, not everyone has the technical knowledge or interest to dive into those configurations


2.The tweak

com.uncore.dig3st_3.jpg
com.uncore.dig3st_4.jpg
com.uncore.dig3st_6.jpg
com.uncore.dig3st_7.jpg
com.uncore.dig3st_8.jpg


2.1.Intro & Support

Dig3st is tested on iOS 15 (on device), iOS 16 and iOS 17 worked fine on simulator, works both on rootless and rootful. Unfortunately i couldn't get iOS 14 simulator running on macOS Sonoma so it's not tested on iOS 14 but from the looks of header files, the support seems to be partial meaning the long look feature may not work on iOS 14, the tweak is very light weight, i only hook three classes.

2.2.What's different than Apple's implementation

  • Not limited to iPhone 15 Pro and above :D
  • Open-source
  • You can use any AI model you like, change the system prompt, user prompt.
  • You can see the full notification (on some apps this works partially for example messages) by long pressing or Apple calls it long looking on it,

3.Setup

Setting it up might be a bit complicated. If you're not sure how to do it, make sure to watch my YouTube video—I covered everything from scratch. I also shared some tips on creating better prompts.

3.1.Model to use

First decide on what model you are going to use, for example Google Gemini has a free tier but they collect data to train their models if it's not an issue for you i'd suggest going with gemini-2.0-flash. If you want to use gemini get your api key from here

3.2.Installation

Install it from my repo, you are going to see links click on the package manager you use and add as source. I have only tested Sileo so far it works fine, if you can't install it from there download the deb file install it manually , here are rootless and rootful debs.

3.3.Change settings

Once you have done those head over to settings app > dig3st > endpoints. If you do not see dig3st there make sure to close the settings app and then reopen. You'll see three endpoints already listed—you may not need to add a new one, long press on the api endpoint you want to edit and add your api key, then click save it's going to push back and save changes.

com.uncore.dig3st_7.jpg


Finally, make sure that the endpoint is activated and enable the tweak from main menu.


com.uncore.dig3st_6.jpg



4.Donate

I don’t run any ads on my website, and this tweak is completely free. Honestly, I believe I could sell it for $3–$4, but I’d rather share it with everyone.

If you like the tweak & if it helped you save time or you just found it useful please consider donating, you can do so from here. If you're unable to donate, you can still support me by watching my YouTube video or starring the repo. Every bit helps, and I truly appreciate your support. Thank you!


5.Build from source

Dig3st was developed on macOS if you are on Linux you might have to do additional steps. All necessary files are available in the repo, after you clone the repo if you have theos setup you can just do

makepackage


The command above will only create a package though if you want to install it your device you have to export an environment variable, make sure to change the ip to your phone's ip

exportTHEOS_DEVICE_IP=192.168.2.100 && make do



6.Why open-source ?

WHY NOT ? DONATE FROM HERE

7.Using local models

If you are thinking of using a local model make sure to use a model that has at least 7b parameters for example i like this one https://ollama.com/library/qwen2.5:7b, depending on the context it feels like there is no difference between gpt4.0 and qwen2.5:7b

8.Debugging (How to test)

Thanks Ayden for saving me a huge amount by having an open-source solution for this already.

8.1.On device

In settings you can navigate to dig3st > settings > test from here you can push fake notifications to test if things are as expected. Make sure that notification content meets the minimum message length requirement.

com.uncore.dig3st_9.jpg



8.2.Using macOS app

This app is brought to you by Ayden. You can test your notifications using this free app only available for macOS unfortunately, get it from my fork download from releases and install . Documentation on how to use the software is available on my fork. Here is how it looks like

notification-debugging.jpg


8.3.Not sumarrized ?

Let's say you pushed a notification or received one and it was not summarized it could be due to

  • The tweak not being enabled
  • The app being disabled
  • An invalid endpoint
  • An invalid API key
  • Notification content not meeting the minimum message length requirement
  • Using Gemini free tier and exceeding the rate limit

You can use the methods above to debug or watch my Youtube video i explained how to.

9.Known issues & improvements in feature

Dig3st works well, but there are a few areas that could be improved. Here are some of them:

  • It does not summarize notification groups
  • If you have set "Show Previews" to "When Unlocked" (meaning notifications for an app are hidden on the lock screen), those notifications are still going to appear. To prevent this, avoid enabling Dig3st for those apps.
  • Toggle full notification works partially on some apps for example messages

10.What's next ?

I am thinking of implementing all Apple Intelligence features, i will probably start with Photo Clean Up feature stay tuned follow my socials especially YouTube


Special thanks to NoisyFlake for Preferences UI (Velvet2) and Ayden for iNDT


Thank you

Comments