DevHeads IoT Integration Server•15mo ago

Hey fellow DevHeads, Sorry for the long absence. I've been working very hard to integrate AI in Embe

Hey fellow DevHeads,
Sorry for the long absence. I've been working very hard to integrate AI in Embeetle IDE:

kristof-at-embeetleOP•10/5/24, 1:16 PM

kristof-at-embeetleOP•10/5/24, 1:17 PM

I'm still in Shenzhen at the moment. Cheers

Kkristof-at-embeetle Hey fellow DevHeads, Sorry for the long absence. I've been working very hard to ...

nour_oud•10/5/24, 1:32 PM

Welcome back! How’s the AI integration into Embeetle going so far? Have you had success with running the LLM locally?

kristof-at-embeetleOP•10/5/24, 2:04 PM

Yeah

kristof-at-embeetleOP•10/5/24, 2:04 PM

As you can see in the video, it is a success

kristof-at-embeetleOP•10/5/24, 2:04 PM

You can select from 15 online and 34 (!) offline models.

kristof-at-embeetleOP•10/5/24, 2:05 PM

The offline models need to be downloaded just once. Then you could plug out your internet connection if you wish so.

kristof-at-embeetleOP•10/5/24, 2:05 PM

Ideal for security

kristof-at-embeetleOP•10/5/24, 2:06 PM

But, I must admit that the local models are a bit slow. The online ones are much faster in responding.

kristof-at-embeetleOP•10/5/24, 2:06 PM

Feel free to give it a try!

Marvee Amasi•10/5/24, 2:18 PM

I recently ran into a tricky memory access violation on x86 64 that wasn't easy to catch with gdb. Switching to Intel's VTune gave me deeper insight into cache misses and pipeline stalls, which made all the difference. In lighter cases, I’ve also used perf to track CPU usage spikes and lock contention across cores.

What tools or methods do you rely on for debugging low level issues on x86 64? Have tools like perf, gdb, or VTune been particularly helpful for you in diagnosing hardware specific issues ?

Kkristof-at-embeetle Hey fellow DevHeads, Sorry for the long absence. I've been working very hard to ...

techielew•10/5/24, 2:41 PM

Is the base AI model GPT?

techielew•10/5/24, 2:42 PM

I'm curious about the development process/experience and what sort of accuracy/consistency you're seeing with it?

kristof-at-embeetleOP•10/5/24, 3:05 PM

There are 49 AI models to choose from

kristof-at-embeetleOP•10/5/24, 3:05 PM

It's powered by "Pieces for Developers"

kristof-at-embeetleOP•10/5/24, 3:05 PM

They build very powerful AI copilots

kristof-at-embeetleOP•10/5/24, 3:05 PM

We integrated their copilot into Embeetle IDE

Kkristof-at-embeetle We integrated their copilot into Embeetle IDE

techielew•10/6/24, 4:44 AM

hmm: https://pieces.app/

Pieces for Developers — Long term memory for developer workflows

Pieces is your AI companion that captures live context from browsers to IDEs and collaboration tools, manages snippets and supports multiple LLMs.

Kkristof-at-embeetle It's powered by "Pieces for Developers"

techielew•10/6/24, 4:45 AM

I wonder how the devs here feel about AI/chatbots/copilots in their development workflows. My casual conversations have yielded all sorts of responses.

techielew•10/6/24, 4:46 AM

We're going to make one available on the site and see how much everyone squeals. I guess that may be the only way.

kristof-at-embeetleOP•10/6/24, 4:54 AM

In Embeetle IDE we provide a chat window that is powered by Pieces for Developers. In there you can discuss with a chatbot about your embedded project (or ask it to generate some code - up to you). The LLM is context-aware: it has access to your project code, so that's the benefit you have compared to just firing up a browser and launching a discussion with ChatGPT over there.

You can then copy-paste things from the Embeetle chat window into your Embeetle code files. That part isn't automated: you need to copy-paste manually if you want to use output from our chatbot. I'm not sure if we even want to automate things here. Doing it manually still feels the safest (I wouldn't want to give up control and let a chatbot mess with my code).

You can select from 49 different LLMs (including ChatGPT 4o, Google Gemini, ...). If you're worried about security, just pick an offline LLM to discuss your code with; then nothing leaks out.

Kkristof-at-embeetle In Embeetle IDE we provide a chat window that is powered by **Pieces for Develop...

techielew•10/6/24, 4:56 AM

Yeah i agree automating it would likely actually slow stuff down by adding the requirement to methodically go through it

Kkristof-at-embeetle In Embeetle IDE we provide a chat window that is powered by **Pieces for Develop...

techielew•10/6/24, 4:56 AM

Which models have you used/tested?

techielew•10/6/24, 4:57 AM

The offline model option is awesome

techielew•10/6/24, 4:57 AM

Curious to hear @ZacckOsiemo's unadulterated thoughts about the above :LolGif:

techielew•10/6/24, 4:58 AM

If you can make Zacck a believer in copilots you have broken the game Kristof

ZacckOsiemo•10/6/24, 4:58 AM

I hate AI

ZacckOsiemo•10/6/24, 4:58 AM

I use it everyday though

kristof-at-embeetleOP•10/6/24, 4:58 AM

There are 49 models to choose from, so I haven't tested them all yet

I must admit that the offline models are a bit sluggish. But that's probably because I'm working on a laptop with a moderate GPU. I guess it will perform better on high-end hardware.

kristof-at-embeetleOP•10/6/24, 4:59 AM

So I prefer to use the online models when working with the Pieces AI in Embeetle. However, I can imagine that this isn't an option if you worry about code leaking out. Then you should pick an offline model.

Kkristof-at-embeetle There are 49 models to choose from, so I haven't tested them all yet 😅 I must ...

techielew•10/6/24, 5:00 AM

What model are you working with now

Kkristof-at-embeetle There are 49 models to choose from, so I haven't tested them all yet 😅 I must ...

techielew•10/6/24, 5:00 AM

@Phoenix

Phoenix•10/6/24, 5:01 AM

kristof-at-embeetleOP•10/6/24, 5:01 AM

@ZacckOsiemo : the AI in Embeetle is not so invasive compared to other IDEs. It's not all the time pushing stuff while you're coding. At the moment it's only present in a separate chat window that you have to open manually. In the future we might integrate the AI in other parts of the IDE, but we'll always be careful not to become "too pushy".

Ttechielew What model are you working with now

kristof-at-embeetleOP•10/6/24, 5:02 AM

I usually work with ChatGPT4-o and Google Gemini. The great thing about the AI chats in Embeetle IDE is that you can switch between models even mid-conversation. So you can try out a bit of everything.

Kkristof-at-embeetle I usually work with ChatGPT4-o and Google Gemini. The great thing about the AI c...

techielew•10/6/24, 5:02 AM

Is there an embeetle API?

kristof-at-embeetleOP•10/6/24, 5:03 AM

Nope.
But there is an API to the "Pieces for Developers" software we used to integrate the AI into Embeetle. It's a Python API.

Kkristof-at-embeetle There are 49 models to choose from, so I haven't tested them all yet 😅 I must ...

Phoenix•10/6/24, 5:03 AM

Have you noticed any significant differences between the models you’ve tested so far? Curious if some are more efficient or better suited for certain tasks.

Kkristof-at-embeetle @ZacckOsiemo : the AI in Embeetle is not so invasive compared to other IDEs. It'...

ZacckOsiemo•10/6/24, 5:04 AM

I like that part, I have to go to a search window to use the AI is a good start