tAI - Get Terminal Commands Using Natural Language

tAI is a simple command line utility tool built on top of Python. Which can generate any commands based on natural language.

What is tAI?

tAI is a minimalistic command line utility tool that integrates AI inside terminal. The purpose of this CLI tool is to generate terminal commands based on english (or other) language.

Example: User says “list all the files in the current directory” and tAI will generate command ls -al and paste it on the terminal. Then user can execute the command by pressing Enter key (there is also a self execution mode).

That’s it! It doesn’t do much or less.

Why I Built This?

I have really bad memory and I always forgot what command I need to pass. One day I was working with gcp sdk and I had to change the current gcp project to another project. And, I completely forgot the command to do so. Then I had to leave my terminal & went to chatgpt website, then asked it, got a response, pasted it in the terminal and pressed enter. This is a huge bottleneck. Instead of leaving the terminal, I should be asking it from my terminal and should get the result. That’s what sparked my first interest in building tAI.

That’s why I built tAI which I can use without leaving the terminal and open source it so that others can also use it.

While I was building I wanted to follow the below design principles:

1. tAI have to be as minimal as possible.
2. It will only do what it was intended to do. Means it’ll only generate bash command and paste it. No agentic task, no file editing, nothing else.
3. It should resemble the AI pop up in the cursor.
4. It should be very easy to setup and use.
5. Nothing should be shared from the terminal for security reason.

How to Use It?

Initially I was targeting only for linux or more precise debian based distribution. But after wasting entire week I was not able to build a .deb file for my package. Then I tried to build the entire python package as binary which also didn’t go well and that’s when I ditched all of those ideas and release it as a python package. Anyone who have python can use it directly. For those who don’t have python they first need to install python. But the python version have to be 3.12 or higher

Simple pip install

The easiest way to use tAI is to install via pip.

pip install tai-textual

Now run tai command and start using tAI directly from the terminal.

This is not recommended. pip will directly install all the package to the global environment and that may caused dependency conflict.

Using Virtual Environment

This is also similar to the above but instead of installing globally this will first create a virtual environment and then install the package inside it. Therefore all the packages (tAI and it’s dependency) will be installed in the virtual environment and will not affect the global environment.

1. First go to a specific directory where you want to install tAI.
2. Then create a virtual environment.

1
2

python -m venv venv

1
2

source venv/bin/activate

.\venv\Scripts\activate.bat

4. Then install the package.

1
2

pip install tai-textual

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

tai() {
    # Path to the virtual environment's activate script
    VENV_ACTIVATE="/path_where_you_have_created_the_virtual_env/venv/bin/activate"

    # Check if the activate script exists
    if [ -f "$VENV_ACTIVATE" ]; then
        # Activate the virtual environment
        source "$VENV_ACTIVATE"

        # Run the 'tai' command
        # The actual 'tai' executable should be in the venv's bin directory
        command tai

        # Deactivate the virtual environment upon completion
        deactivate
    else
        echo "Error: tAI virtual environment not found at $VENV_ACTIVATE"
    fi
}

After pasting this function inside your .bashrc file, you need to change the VENV_ACTIVATE path to the path where you’ve created the virtual environment. Then, restart the terminal or run source ~/.bashrc to apply the changes. Now you’re all set! Just type tai in your terminal to use tAI.

Installing via pipx

This one is for Linux or Ubuntu user. It is not recommended to install globally via pip in a debian based distribution. This can corrupt the entire system. This is where pipx comes into the picture. It creates an isolated environment for each package and installed them there. Instead of creating a virtual environment manually, installing via pipx can be handy.

1. Install pipx:

sudo apt install pipx

2. Install tAI via pipx:

pipx install tai-textual

That’s it! Now you can use tAI by running tai command in the terminal. You don’t need to worry about activating the virtual environment. pipx will handle everything.

What Features tAI have?

Though the principle of tAI was to be as minimal as possible. But I’ve to balance between minimalism yet configurable. That’s why I’ve added many customization features.

21+ LLMs

Currently tAI supports 21+ LLMs from different provider such as Google, OpenAI, Anthropic and Open router for free models. User have the flexibility to choose any of the models they like. They can change the model from the terminal utilizing terminal user interface (thanks to textual). I’ll be adding more models in the future.

Free Usage without Seting API keys

You’ll have access to Free LLMs from the get go. You’ll install the package and start using it without seting any api keys. Currently tAI offers total 7 open source models completely free to use.

Default LLM

Set a default LLM so that everytime tAI starts it is set to the default model. You can set the default LLM either from the settings tab or using command.
Command to set default model:

1
2

tai --default-model openai/gpt-4o

Inline & Fullscreen Mode

tAI have 2 screen modes. One is inline mode similar to cursor, another one is fullscreen mode. This gives the user complete flexibility to use any of the mode as per his liking. Changing between this two mode is also very simple. Either can use direct command (tai --fullscreen true) to change between fullscreen and inline or use the settings tab to change the mode. You can find this under the others section from the settings.

It is advised to use the fullscreen mode while using tAI inside tmux.

Auto Paste & Execution Mode

There is 2 mode. One is Auto paste and another one is Auto execute. User can toggle between this mode using ctrl+e keybind. In the Auto paste mode tAI generates a command and paste it to the terminal. Then user can see the command and decide if he wants to run it or not. Meanwhile Auto execute model runs the command itself. My advice is to use the Auto execute mode with caution. Cause LLM can generate a command which can delete system files. You don’t want to blindly trust LLMs.

Custom Prompt Config

To make it more customizable and more personalized, I’ve given the user the choice to customize the system prompt. User can customize the system prompt from the prompt config section inside settings. You can guide the LLM to your personal need. This is completely upto the user. Also you can see the default system prompt from this prompt config section.

Seting API keys

You have free access to LLM. But it is still a better choice to bring your own API key (paid or free) for better experience. Free LLM will have rate limitation and slow response time. For the best experience bring your own api key. Currently tAI have integration with 4 providers:

1. Google (Gemini models)
2. OpenAI (GPT models)
3. Anthropic (Claude models)
4. Open router (for all model)

You can set any of the provider’s API key from the settings tab or using the command line.

There is another option in the others tab of settings. Where you can set all the models to use Open router api key. Then you only just provide your own Open router api key and use every model (OpenAI, Google, Anthropic, etc). But then the default open router api key (which was being used for free models) will be disabled.

Future & Contribution

This is an open source project so I can’t make any promises but I’ll try my best to add more features or improvement to tAI. At this moment I’m working on the following features:

1. Add more models (will be adding new kimi k2 model)
2. A docker image (pull the image and start using without any hassle)
3. Add context (currently there is no context added to LLM call, if it makes mistake it won’t know, will be adding context so that it knows about previous query and response)

But, I welcome any contribution to this open source project. If you encounter any bug then create an issue, or ask for a new feature or create a PR with new feature. Below I’m adding the project repo.

1. Github: tAI
2. PyPI: tai-textual

Special Thanks

This project is build on top of 2 great libraries.

1. Textual: This framework was the MVP. Everything you see in the terminal is build on top of this library.
2. LiteLLM: This is an API wrapper for all the LLMs. It made my life so much easier.

Connect With Me

I’m building stuff with AI/LLM. If you’re interested in seeing my work or collaborating, feel free to reach out on my socials.

1. Facebook: Shoaib Hossain
2. LinkedIn: Shoaib Hossain
3. X: @KillerShoaib__

And I have also started writing blogs about my work related to AI and other stuff. If you’re interested, then check out my blogs.

• My Personal Blog: AI With Shoaib

See my opensource work from Github.

• Github: KillerShoaib