Hello qwersyk,

I'm writing to let you know about a problem I'm having with the Newelle app on Vanilla OS 2 Orchid.

Your application is extremely useful for Linux users, providing a wide range of functions, including file creation, system management, and Python code execution. However, I ran into a bug that prevents me from sending messages to the bot.

I have attached a screenshot of that error. I would appreciate it if you could investigate this issue and implement a solution as soon as possible.

I would also like to suggest better integration of Newelle with Vanilla OS 2 Orchid. The documentation available at https://docs.vanillaos.org/ could be a valuable resource to facilitate this integration.

I strongly believe that Newelle has the potential to be an exceptional virtual assistant for Linux users. With its unique extensions and advanced features, Newelle differentiates itself significantly from other available options.

I would like to request that you consider the following improvements:

• Update Prioritization: Ensuring periodic updates to keep Newelle compatible with the latest versions of Vanilla OS.
• Optimization for Vanilla OS: Adapting Newelle to the specifics of Vanilla OS 2 Orchid, providing a smoother and more integrated user experience.
• Implementation of Advanced Commands: Extending Newelle functionality to include Vanilla OS specific commands such as power management, network configuration and peripheral device control.

I am convinced that implementing these suggestions will help strengthen Newelle's position as an essential tool for Linux users, providing them with a complete and intuitive solution for interacting with their operating system.

Thank you for your time and attention. I look forward to solving the problem with sending messages and implementing the suggested improvements.

Regards,
Raul Popescu

Many models support images and files. We should probably implement a way to do it.

We should add Text To Speech (I would say disabled by default), offering different programs/APIs and languages to choose from as for user preference.
Some of the services/programs we may want to include:

• ResponsiveVoices
• espeak
• festival
• VoiceVox
• gTTS

From some programs, it probably is better to just call them outside of flatpak rather than adding more dependencies

Prompt

"List out all the PDFs in the Downloads folder"

Expected behaviour

A list of PDF files from the Downloads folder

Actual behaviour

App crash.

Not sure if this is helpful, but I see the following in the terminal as the app crashes

(newelle:2): Gtk-CRITICAL **: 02:32:29.828: gtk_text_attributes_ref: assertion 'values != NULL' failed
**
Gtk:ERROR:../gtk/gtktextlayout.c:1261:set_para_values: code should not be reached
Bail out! Gtk:ERROR:../gtk/gtktextlayout.c:1261:set_para_values: code should not be reached

We can use SpeechRecognition library that supports a lot of services and looks pretty easy.
To enable speech recognition, we can add a microphone icon to hold or just click near the message box. It would be also nice to add something like keyword recognition to activate it.

There is actually no big reason to need an app restart to update the settings. I guess we can just apply them when preferences window is closed by updating settings variables

This is an amazing application but it would be so much better if there would be the capability of running locale models, like this censoring is not a problem but mainly "openai" is not getting lists of files and so on.

Hello,

I just discovered this project and I think it's really cool.

The ability to run commands using the flatpak sandbox is really nice, but do you think it would be possible to have the option to see what command will be run in the console before clicking on a button to validate/cancel it?

It might be nice for people running commands without a sandbox, or just to check what is happening with the model (I'm sure nobody likes to get a rm -rf / or a fork bomb prompt-injected into their console, even in a sandbox).

Thanks again for the hard work!

Last year there were problems with streaming with GPT4All so we didn't implement it. Now they should be solved.
We should implement it

Since some services support it, it would be a nice thing to add, specially for local language models that generate text in a pretty slow way

Having fixed prompts causes these problems:

• Some AI models we may want to add in the future support system prompts
• Being the text almost all in English, often the assistant starts speaking it even if most of the conversation was in another language
• If the user wants the AI to behave in a certain matter, it will often don't

Also, it would be just nice to have the user customize the prompt.
How I think we can integrate it:

• We store the prompts in a json saved in a dconf string setting. I think that it is more clever than adding multiple dconf entries.
• We add an actionrow in the prompt control section of the settings with a button that opens another window where the user can customize the prompts text

I think the latter is the best way since settings are becoming a bit too long and may overload the user with information.

This issue is probably caused by some place where we specified .var directory bad, since it creates a .var/...etc. directory in the folder where Newelle was launched

At the moment, generating a title for a specific chat will always make a request with the current history

Suggestion/Bug Fix

Proposal : Implement a system to allow BAI Chat to be connected to via an API Key.

Explanation : This would fix the current issues with BAI Chat and open the way to changing the BAI Chat model in the future.
I am currently unable to find the name of the BAI-Chat free model (and I am unsure if it even exists) in the BAI Chat documentation (https://docs.theb.ai/quickstart) but believe that adding API functionality will be beneficial in the future, particularly if we also were to implement the ability to choose the model.

(I will be trying to learn how to implement this myself but it will be a long time before I can manage such a thing. Thank you for making such an amazing project!)

There are many hardcoded path in this app, which works only with snaps. Kindly generalise them and enable support for other packaging formats like snap. Patching them is often tough, cause we need to update the patch also with each release. Also, the flatpak spwan command is used which is unnecessary for Snaps.

It would be nice to show the user a small presentation that explains Newelle's features and capabilities to the user. It should also show the prompt to give Newelle all the extra permissions to operate on the user system.

Hello,
I just discovered this app as a flatpak and it is awesome !
But unfortunately, the app just crah when I type certain prompts. For example :
list all files in /home/darkpepito result in a crash.
I use i3 as my main WM but I don't know if it's important.

Here are the specs of my machine :
Dell Latitude-E5470
CPU: Intel cored i5-6300U
GPU: Intel HD graphics 520
16 GB of RAM
OS: Zorin OS 16.2

Thank you for your time !

Installed with Flatpak and it is not working "Failed to send bot a message"

Am I missing something? Thanks.

It would be a good idea to add support for local language models.
We can think about using GPT4All bindings for python that looks pretty easy to use.

Also, we should add a voice in settings to make the user choose which language model to use, which prompt to use, and probably also a "download manager".

When trying to use Newelle, i receive "failed to send bot a message"

Installed on Linux on ChromeOS, that chat is deadly white, the menus seem fine.

Can you please try to fix it? That would be worth a coffee.')

This will allow the file picker to see extensions without needing read permissions.

ok so i am in a debian vm.
i ran this promt:

it worked without any problems.
but than i got this without typing anything:

and this is not the user and the home folder.
Maybe i got the output of someones other promt?!

Since Bai Chat now requires login, it can't be used anymore as default provider. We should find another provider that does not need additional steps to configure. There are many providers we can use from Gpt4Free, however, I'm not able to install g4f on build.
Alternatively, we can think about integrating one of those by ourselves or find another library that is easy to build

Loading custom models in GPT4All is possible, here is the guide.
We should implement a way to do it.