I personally use it to track the tides and get notified when certain conditions are met, something that pure LLMs had trouble dealing with and pure code was often too brittle for.

Happy to answer any questions folks have!

As an example in Javascript, simply calling `throw new Error('break!')` will stop the run from continuing.

We plan to add conditionals, branching, etc. Baby steps :)

Most of the code is actually pretty simple, the complexity comes from the code runners and the infra needed to support the various integrations.

To open source it, we'll need to create some local code-runner alternative, as not everyone wants to spin up AWS resources. With that said, we're excited about running this locally ourselves (e.g. scripting various things on-device).

Right now the code blocks can't access the internet, but the possibilities will be endless when that lands.

Would love any integration requests!

(though as it still involves LLMs I'd probably want to use it for stuff that sounds less mission critical than some of your examples!)

We see a future where the code can be "self-healing" as well, with user approval of course.

This may be useful but would love some clarity on how long you're gonna keep it like this.

For those finding the service useful already: we're also thinking about open sourcing it so you can just run the loops yourself to keep costs down.

whisper voice input ("hey, could you tell me what alarms I have tomorrow?") ->

LLM command interpretation ("match input query to pre-defined commands and arguments", "LISTALARMS, , , 8, 2, 2023") ->

underlying script ("{'23:00', 'Doctor's Appointment'}") ->

natural re-statement ("re-word this data as an answer to the question xxx", "you have a doctor's appointment tomorrow at 11 o'clock PM") ->

so-vits/tortoise-tts/some TTS plugin

nodes could be fabbed out of single-use .py scripts, and the whole thing could be compiled into a headless script. endless possibilities.

It turns out that when you write simple programs, they often resemble what you'd find on Zapier or IFTTT :)

https://news.ycombinator.com/item?id=36524078

I really hope you open-source this :)

Is there no way to bypass this currently?

Second, this is what I remember Retool looking like long before it became a huge tool. I feel that this is a step towards building bespoke tools using LLM tech. (Of course, I'm sure others started in similar ways, but this is one I remember.)

File "/home/glot/main.py", line 1 var BLOCK_INPUT = `"If you could instantly become an expert in any one subject, what would it be and why?"`; ^ SyntaxError: invalid syntax

Each block is editable, even if it doesn't seem so at first (including output!)

It's worth noting that this is not AutoGPT -- the initial generation is just a starting point, and that the actual loop configuration and validation is entirely within your control.

Suggestion: please add the ability to use credential for social, to be able to perform social tasks.

Our current plan is to keep Magic Loops running as long as people seem to like it (and especially because I track the tides with it!), but we also think it could be super powerful to run locally, so we're hoping to open-source it at some point as well.

Will definitely play around with it more

Also, we hope that an open-source version will allow folks to do more with their own keys.