Still very much working on it 
I realized that to make this different from other AI assistant apps, I need to integrate it deeply into BetterTouchTool and make it really flexible. This turned out to be more complex than I initially expected.
Currently testing a initial version with some friends. It is almost ready for deploying it to all alpha users but a few important changes are still planned for the next few days... I'll post an update here soon.
@Andreas_Hegenberg Can't wait. You could also look at the UI/UX of this app. It's built by the developers who created the Workflows app; Apple required it and it became Shortcuts.
I was just using BTT and was thinking about AI features, good to know you're already working on it! This is what apple intelligence was supposed to be!
Please take you time, sloths must not be rushed!!! 
i never thought about llo looking like a sloth, now I can't unsee that!!! Very coooll
hi andreas. is it available in alpha ?
will it be a separate install, or directly baked in in BTT? i have to say it looks powerful and amazing. are you the first to do such a project ?
the alpha contains the basic building blocks (ai-chat, tools, sub-agents), but I'm gone for the week and will only be able to put everything together next week. I will share a sample setup here.
Curious if it only supports local and OpenAI at the moment?
I've tried adding Claude - Model identifier "claude-opus-4-20250514" and endpoint https://api.anthropic.com/v1/chat/completions
However, in the api area there's only an "Open Ai Api Key" available so it seems like OAI, unless I'm missing something?
Thanks!
Ok, user error. I saw switch to other models.. and MCP etc. This is pretty wild
Mind-blowing! Can we use Google Gemini's API for it? I tried but didn't find a way to make it work
gemini also works, but the current alphas are not fully ready. I’ll now make everything I have so far available on Monday with the Thaoe release of BTT! Currently working on the docs.
API URL for gemini should be https://generativelanguage.googleapis.com/v1beta/openai (openai completions compatible)
it's not yet available though is it ?
Hi @Andreas_Hegenberg, from your experience, which LLM models produce the highest quality output? I'm also curious if you've found a model that produces surprisingly good output given its relatively lower price compared to more powerful models.
Following
Did you test GPT-5? It requires ID verification first. :)) I'm using Gemini 2.5 Flash and I like it.
I'll soon do some summaries of my findings and share some of my example agents! GPT-5 is good, but also pretty slow.
Small UI & UX comment. In Dark mode, some text is difficult to read due to low contrast with the background. Specifically, the purple colored text without any background pill.
You can modify some of the colors in the appearance settings of the assistant, I'll soon make all colors configurable
I think these use the same color as the user bubbles currently
Hi @Andreas_Hegenberg , I found a bug for "Window & Display" Tools. From my usage, it appears that any function that accepts an id argument will crash BTT.
I figured out that h@llo.ai will try to execute ai.screenshot_window or ocr_window and always crash.
To show you what I mean, try running this JS, you'll see that it crashes BTT:
//if you rename this, make sure to also rename it in the "function to call" field below.
async function someJavaScriptFunction() {
let windows_array = await ai.get_windows_sorted_front_to_back({numberOfWindows: 20});
let first_window_id = windows_array[0].id;
let screenshot_result = await ai.screenshot_window({id: first_window_id});
return screenshot_result
}
thank you I'll check - I believe I renamed the parameter but forgot to merge something
without configuring anything it looks weird/very transparent/unreadable on macOS 15:
I’m very interested in where this is going, BTT is surely one of if not the best tools for macOS =)
the transparency issue on macOS 15 is already fixed in the latest build.
The screenshot function should now work with 5.634
There will be quite a few updates to this in the coming weeks, I also haven't deployed all of the RAG services with the BTT documentation. Once that is done it will be able to configure most BTT triggers for you.
