DEV Community: ColeDrain

ChatGPT in the Shell

ColeDrain — Sat, 25 Mar 2023 14:17:37 +0000

Introduction:
Hello there amigo 👋.

In this tutorial, I will show you how to create a CLI app that allows you to interact with ChatGPT from the command line. The app, named "ChatShell", will allow users to ask for shell commands for specific tasks or chat with ChatGPT directly. We'll be using the Go programming language and Cobra CLI library to create our CLI app and the go-openai library to interact with OpenAI's GPT-3 API.

Please do give a thumbs-up, if this was insightful in any way.

Prerequisites:

Go programming language installed (version 1.20.2 or higher)
Cobra CLI library installed
go-openai library installed
Viper library installed (for config and environment vars)
An OpenAI API key

End Goal:

Able to run shell command of form:

chatshell ask "how to delete a local branch ?"

chatshell ask -c "when was gpt first built ?"

Steps:

A. Set up the project structure

Create a new directory for the project and navigate to it

$ mkdir chatshell
$ cd chatshell

Initialize a new Go module

$ go mod init github.com/{your_username}/chatshell
# note: replace {your_username} with actual github username

Install the Cobra CLI library and go-openai library

$ go install github.com/spf13/cobra-cli@latest
$ go get -u github.com/sashabaranov/go-openai
$ go get github.com/spf13/viper

B. Create the CLI app

cobra-cli helps generate boilerplates for your cli app automatically, this speeds up development.

Use the Cobra CLI to generate the basic structure of the CLI app:

$ cobra-cli init

This will generate the basic structure of the CLI app, including a main.go file, a cmd directory, and a LICENSE file.

You can have a look at the code if you're familiar with cobra.

C. Implement the 'ask' command

Once again we will call on our friend (cobra-cli)

cobra-cli add ask

This will generate a file ask.go which will contain the implementation of the 'ask' command. We have to finetune the boilerplate to our need, below is finetuned code well commented for understanding.

Import necessary packages

/*
Copyright © 2023 Ugochukwu Onyebuchi <pyvinci@gmail.com>
*/
package cmd

import (
    "context"
    "fmt"
    "os"
    "runtime"

    "github.com/spf13/cobra"
    "github.com/spf13/viper"
    openai "github.com/sashabaranov/go-openai"
)

Next intialize the askCmd
Use: a short description of how to use the ask command, in our case we expect ask to be followed by a text in quotes
Short: a short description of the ask command
Long: a longer description
Args: we specify we only want one argument, which is a text in quotes
Run: we specify the function to be called, this function will handle the business logic.

// askCmd represents the ask command
var askCmd = &cobra.Command{
    Use:   "ask [text]",
    Short: "ask ChatGPT",
    Long:  `ask is used to ask for shell commands for a particular task`,
    Args:  cobra.ExactArgs(1),
    Run:   runAsk,
}

// authToken holds your openai auth key
var authToken string

// initial operations
func init() {
        // add the ask command to root
    rootCmd.AddCommand(askCmd)

        // add flag to enable chat mode
    askCmd.PersistentFlags().BoolP("chat", "c", false, "Chat with ChatGPT")

    // Read the auth token from a configuration file
    viper.SetConfigName("config")
    viper.AddConfigPath("$HOME/.chatshell")
    viper.AddConfigPath(`%USERPROFILE%/.chatshell`)
    viper.SetEnvPrefix("OPENAI")
    viper.AutomaticEnv()

        // Handle missing config file or auth token
    if err := viper.ReadInConfig(); err == nil {
        authToken = viper.GetString("OPENAI_AUTH_TOKEN")
    } else {
        fmt.Printf("Error reading configuration file: %v\n", err)
    }

    if authToken == "" {
        fmt.Println("Error: OPENAI_AUTH_TOKEN environment variable not set")
        os.Exit(1)
    }
}

The prompt I use is a bit large: but for much correctness I get the osType using runtime.GOOS and SHELL type using os.Getenv("SHELL"), this will help ChatGPT give reasonable shell commands of the relevant OS, we don't want shell commands for mac whilst we are using windows.. and it also responds with "Sorry I can't seem to find a command for that", if it can find a command for your prompt. It took a while to get a good prompt as this, it's an iterative process..

// Call OpenAI and prints response
func runAsk(cmd *cobra.Command, args []string) {
    client := openai.NewClient(authToken)
    osType := runtime.GOOS
    shell := os.Getenv("SHELL")
    content := ""

    chatMode, _ := cmd.Flags().GetBool("chat")

        // We use the state of the flag to determine the context
    if !chatMode {
        content = fmt.Sprintf(`You are a very helpful shell assistant that gives users only shell commands to achieve a task, just give out only the shell command(s), and nothing else, no preamble, greetings or explanation please, just the shell command. When you can't find a command for a query/prompt/greeting respond strictly with "Sorry I can't seem to find a command for that". Start now: "%v in %v os using %v shell"`, args[0], osType, shell)
    } else {
        content = args[0]
    }
        // Get ChatGPT Response
    resp, err := client.CreateChatCompletion(
        context.Background(),
        openai.ChatCompletionRequest{
            Model: openai.GPT3Dot5Turbo,
            Messages: []openai.ChatCompletionMessage{
                {
                    Role: openai.ChatMessageRoleUser,
                    Content: content,
                },
            },
        },
    )

    if err != nil {
        fmt.Printf("ChatCompletion error: %v\n", err)
    }

    fmt.Println(resp.Choices[0].Message.Content)
}

D. Update the configuration file
For all this to work you will need your OPENAI_AUTH_TOKEN set.
Remember, we set viper to look for a config file that should contain our OPENAI_AUTH_TOKEN

Create a new directory called .chatshell in your home directory (on Windows, use %USERPROFILE% instead of $HOME):

$ mkdir ~/.chatshell

Create a new file called config.json in the .chatshell directory and add your OpenAI API key as the value for the OPENAI_AUTH_TOKEN variable:

{"OPENAI_AUTH_TOKEN":"your_api_key_here"}

E. Build the CLI app
Navigate to the project's root directory and build the app:

$ go build

This will create an executable binary called 'chatshell'.

F. Use the ChatShell CLI app
Run the chatshell binary and provide a task description as an argument:

$ ./chatshell ask "how to delete a local branch ?"

output: git branch -d <branch_name>

The app will then interact with ChatGPT and return the shell command for the given task.

You can also use the -c flag to chat with ChatGPT directly:

$ ./chatshell ask -c "When was gpt first built?"

output: GPT (Generative Pre-trained Transformer) was first introduced by OpenAI in June 2018

Conclusion:

In this tutorial, we saw how to create a Conda CLI app that interacts with ChatGPT from the command line. The ChatShell app can be used to obtain shell commands for specific tasks or chat with ChatGPT directly. Feel free to explore the provided code and the go-openai library to extend the functionality of the app or adapt it to your needs.

Repository: https://github.com/ColeDrain/chatshell

Search Videos/Audios Easily

ColeDrain — Tue, 25 Oct 2022 12:44:53 +0000

A while back I had the issue of figuring out where a particular thing was said in a video or audio, so I thought "was it possible to search a video/audio ?".

I knew what the steps would look like (Transcribe -> Parse/Search Transcripts), but it was kinda hard to get really good accuracy on transcription, no that wasn't really the problem, the problem was I needed an open source STT model I could easily iterate upon freely and also wrap my head around, I tried some cloud STT/ASR's but I lacked that freedom. Then came OpenAI with Whisper, the accuracy is really good even with the tiny.en model, then another issue I had was deploying the model and creating a UI for interaction, then came Streamlit, it was really fun creating with Streamlit's widgets and then deploying on Streamlit cloud, smooth as butter.

Finally, I came up with this: Search Media Web App, I'm curious what people think about this, I really don't know what to think of it, but I'm glad I was able to solve a problem I had, so send in your feedbacks or just try it out..

Github Repo
Whisper Repo

ML-Zoomcamp Series (wk 2)

ColeDrain — Tue, 20 Sep 2022 00:31:30 +0000

Hello there, this is week 2 on my Machine Learning journey with ML-Zoomcamp, well this week was pretty choked up, but I still made time to go through the content, here are some things we did this week..

We were introduced to a real world problem, that involved predicting the price of a car.
First, we had to gather, access and clean the data to make it good for the ML model.
Explored the dataset visually through graphs and chart to spot outliers and understand the distribution of target variable.
We then split the dataset into Train, Validate and Tests sets.
Linear regression model is a very popular model for predicting the price of a numerical quantity or variable, we used this, and we implemented the linear regression model with numpy.

ML-Zoomcamp Series

ColeDrain — Mon, 12 Sep 2022 08:03:22 +0000

I Recently joined ML Zoomcamp to build my ML skills.

These are some things I have learnt, I hope you could learn something too.

Week 1:

What's Machine Learning ?

Machine learning is the ability of machines to pick patterns from data and its features, and use the patterns to make predictions to new data.

g(x) ≂ y

ML VS Rule-Based Systems(RBS)

RBS systems involve you identifying every possible scenario and creating a rule for such, this is a very tedious process and is very limited and time-consuming, these rules can instead be used as features to a ML model. The model is trained on a collected data with these features.

Supervised ML

In this type of ML, a model is trained on features (x) and their corresponding target variable (y) or labels.

CRISP-DM (Cross Industry Standard Process for Data Mining)

CRISP is a methodology for organizing ML projects, it's been around for a while now, old but still used. It consists of 6 major steps as shown below..

Model Comparison Problem

When comparing models in your evaluation, be careful, as models can get lucky, so it's advisable to also perform validation and test on each of the models you're comparing to make the best choice.

Train -> Validate -> Test

Also had a numpy, pandas and linear algebra refresher. Then was given an assignment to reinforce all this.

You can also follow via this repo Ugo's ML Zoomcamp

Starting an Internship at HNG - My Goals Checklist

ColeDrain — Sun, 15 Aug 2021 13:58:09 +0000

Hello devs , one thing I have learnt in my dev journey is the importance of having a goal, so this is me documenting my journey through the HNG internship.

A quick one about HNG, HNG is a virtual internship operated by Zuri, that helps devs scale their career with real world experience.

I believe setting goals for any journey, can help you appreciate how far you've gone at the end.

I would love to push myself a bit, so I'm participating in 3 tracks: Backend, Frontend and Mobile. I know it's gonna be tough, but I believe I can do this.

My Goals include:

Create meaningful connections with other developers.
Improve my tolerance to pressure.
Improve my current know-how in the back-end, front-end and mobile stack by learning from mentors.
Create projects confidently from start to finish.
Learn how to work with a team(never have).
Gain real world experience

I will be back here again to see how well I did.

Wish me luck

For more details on HNG visit Zuri Internship

Plus:
Some Resources to Help Beginners

Chatting Via the Command Line (ssh)

ColeDrain — Wed, 07 Jul 2021 16:33:43 +0000

Hello devs, trust you're doing great.

I came across this cool repository on GitHub Made by Ishan Goel.

It's called Devzat

Devzat is chat over SSH: devzat (GitHub Link).
Because there's SSH apps on all platforms, even on mobile, you can join from anywhere.

To use, just open your command prompt and run this command:
ssh devzat.hackclub.com
make sure you're connected to the internet, of course.

Run cmds to see a list of commands.

Interesting features:
• Rooms! Run cd to see all rooms and use cd #foo to join a new room.
• Markdown support! Tables, headers, italics and everything. Just use \n in place of newlines.
• Code syntax highlighting. Use Markdown fences to send code. Run eg-code to see an example.
• Direct messages! Send a quick DM using =user or stay in DMs by running cd @user.
• Timezone support, use tz Continent/City to set your timezone.
• Built in Tic Tac Toe and Hangman! Run tic or hang to start new games.
• Emoji replacements! 🚀 => (like on Slack and Discord)

Devs this is cool, come try it out, and drop your feedback.

How I ran Django on my android phone ❌💻

ColeDrain — Wed, 02 Sep 2020 11:54:48 +0000

How I ran Django on my android phone...

Yeah, it seems weird coding with your phone, but right now I don't have an option, so if you are like me or you just fancy the idea, you can also do this. This is not an introduction to Django tutorial but how to work on Django with your android.

First we need something like a Linux terminal(or CLI) for android. Lucky for us, there is a terminal emulator on Google play store called Termux, so head to playstore and install it.
Link:termux

Okay, so you have installed it, what next?
So basically you can do whatever you could do on a Linux terminal now with this app(create directories, move files, change directory). This is how Termux looks like.

So let's roll in. Enter this command to view where Termux saves your files.

$ pwd

So that's where Termux saves your files, but the problem is you can't access that folder because of safety issues.

So to be able to access your files..
run this command

$ termux-setup-storage

And allow permission to storage in the popup shown.

Now change directory to your internal shared storage by running this command.

$ cd /sdcard

N:B this command doesn't point to your external SD card but your internal SD card(storage).

Then let's create a folder to store our Django projects, let's call it django_projects.
So run this command.

$ mkdir django_projects

$ cd django_projects

Now let's install python
To install python, run this

$ pkg install python

To check if python is installed, run the python command to enter python console.

$ python

N:B Use ctrl + z to exit the terminal. Termux app comes with a ctrl key.

Let's create a project called mysite
So still in that django_projects, run this commands.

$ mkdir mysite

$ cd mysite

We need a virtual environment to avoid dependencies clash. So let's install a virtual environment, run this command.

$ pip install virtualenv

After installing it, run

$ virtualenv venv

to create a virtual environment.. Then activate the virtual environment by running (note the full stop)

$ . venv/bin/activate

Now we have a virtual environment, let's install Django. run..

$ pip install Django

Okay, so Django is installed in our venv, let's create a Django project. run..

$ django-admin startproject hello .

(note the full stop)

Then let's run migrations..

$ python manage.py migrate

You can then run the server..

$ python manage.py runserver

voilà, go to your phone browser and open http://127.0.0.1:8000 and you should see this.

You can always view your files by going to your file manager and searching for django_projects folder.

Feel free to indulge me, if any issues.

Social Authentication or Not?

ColeDrain — Sun, 05 Jul 2020 13:03:13 +0000

Good day devs.

I have been thinking about this, and I want to know what you think, or your experience with it.

I basically want to create a user authentication system, do you think social Authentication is okay or should I do a basic registration->login flow. I am mindful of issues that may arise in the future.

Thanks, I would really love to hear from anyone.

Wanna Start a Journey Into Tech Writing, get in here.

ColeDrain — Sat, 13 Jun 2020 10:02:04 +0000

One thing I found out is, some times it becomes hard to start something, when you see so many people that are 'big names' or experts in that field you want to enter. It's like a standard is already set, and you feel what's the need of even starting.

You forget that those people ahead always had a beginning, and like you they likely had people ahead of them.
So friend don't be afraid to start, just do it.

Some Useful Tips for Tech Writers

Have the interest & Choose a Niche

This cannot be overemphasized, else you wane out in the journey, choose a niche you find interest in or are curious about.
There could be so many things that drive your passion for writing, but overall let it be that "you are delivering a great content to the world" that people will enjoy.
There are many areas in tech you can write about, though it's possible to be a 'Jack of All Trade', but I wouldn't recommend this for someone still new to writing in the tech world.

If you find choosing a niche difficult:

You can check out what others are writing about, along the line you may get inspired.
You can also write about your learning journey no matter what level you are in, it will help others.
You could also choose a niche based on what's trending, what people are searching about, just go to Google Trends, you will see what people are searching about.
Some technologies that people are not so clear about, you can do research, then break it down in a clearer way to people. and many more.

Have mentors or (people who also write)
On dev.to for example, you can follow writers who you enjoy reading their content or are also in your niche. You could also have a mentor who will guide you, there are a lot of good writers out there who you could learn one or two from.

Do your research well
When writing have it at the back of your mind, that people are definitely gonna read your content, and it would be bad if your content contains error or it's outdated, it may be frustrating on the reader's end.
Your content is what always brings the user back, if you want to make content that brings readers back then do your research well. Remember great writers are also great readers.

Other tips:
Be specific- let's say you are writing a tutorial for Django developers, specify the technologies and tools you are using, it will prevent some 'banging of the keyboard moment'.
Feedback- You can always ask for feedback and do well to help readers who are confused about something in your content. I have seen many contents on the web that lack this, they just drop content for the readers and that's it.
Keep Yourself updated

Last, the ocean you see today was formed by yesterday's little drop, add a little drop daily and you will be surprised at the ocean you form, write as much as possible and enjoy yourself.

This post first appeared on ugoayo.pythonanywhere.com

Web Monetization API, What you need to know. Short and concise.

ColeDrain — Fri, 12 Jun 2020 22:32:54 +0000

Web monetization in layman terms is making money from traffic to your websites, as easy as that. Most websites do this through:

Pay Per Click PPC- get paid when an ad is clicked by site visitors
Cost Per Impression CPI- get paid when an ad is displayed and viewed by site visitors.

So basically the existing ways of making money from your website involved putting ads on your website, and truthfully ads atimes are very annoying, your privacy is invaded, user experience is diminished for example why will an ad cover the content of a web page you are reading, it's so annoying.

A new standard of making money on the web is being advocated, it is called the Web Monetization API. Web monetization API seeks to solve the problems of present monetization schemes, that is no ads, no invasion of privacy, the user enjoys the website, and the website owner still earns.
From webmonetization.com, web monetization API is a JavaScript Browser API which allows the creation of a payment stream from the user agent to the website, the payment occurs in real time. Just by adding a line of code to your website or specifying a streaming channel to earn.

How it works:

The website or website owner needs a wallet to receive payments.
User/ Website visitor needs a wallet to send payments.
The browser initiates payment in the background.

Chrome and Firefox browser support it as an extension.
Puma browser supports it totally.

So a future without annoying ads and privacy thieves may be possible. Keep this in mind.

Check out the following for more gist:
webmonetization.org
coil.com

This post first appeared on ugoayo.pythonanywhere.com