A command-line tool that allows you to interact with GPT-4 directly from your terminal, including long form conversations via chat

To use ShellGPT, simply run the command with your input message:

gpt "What is the meaning of life, the universe, and everything?"
# ... often humorously said to be the number 42 ...

You can also run shell commands with -x:

gpt -x "Undo my last git commit"
# git reset HEAD~1
# Are you SURE you wish to run the above command? (y/N): y

gpt -x "Download the wikipedia page for Capybaras with all images"
# wget -r -l 1 -H -t 1 -nd -N -np -A jpg,jpeg,gif,png -erobots=off https://en.wikipedia.org/wiki/Capybara

And pipe input and output, transforming files and commands (pass - to accept stdin):

cat mod.ts | gpt "Generate a helpful README file for this project" - > README.md

cat crash.log | gpt "Why is this crash occurring?" -

gpt "Output a CSV of 10 notable cities in Japan with their name in English & Japanese plus a fun fact enclosed in quotes" > japan.csv

Chat with gpt from the terminal

gpt --repl

> hello there
Hello! How can I assist you today?

>

1. Install the Deno runtime.

2. Run the following command to install ShellGPT:

deno run -rA https://deno.land/x/shellgpt/install.ts

3. Ensure $HOME/.deno/bin is in your PATH.

export PATH="$HOME/.deno/bin:$PATH"

You can now use gpt [my request] - or whatever you named the script during installation

That's it!

Note: If you add that export line to your .bashrc or .zshrc (or equivalent), remember restarting your shell to get the updated PATH

You must set the OPENAI_API_KEY environment variable in your current shell, using a key obtained from https://platform.openai.com/account/api-keys:

export OPENAI_API_KEY=...

You can also set OPENAI_CHAT_URL if you'd like to target a different OpenAI-compatible API endpoint (e.g. Local Llama, Anthropic, etc.).

To configure the specific ChatGPT model, system prompt used, and more, you can type gpt --config

These commands are used for general ShellGPT-wide operations.

These commands are for specific chats, either new or existing.

These commands perform non-textual tasks.

Shell-GPT has some useful and unique features:

• Execute shell commands with a confirmation step (just pass -x).
• Supports input/output piping for simple file creation and transformation (see Basic Usage).
• Utility commands for convenient chat history viewing and editing.
• Smooth, streaming output, resembling human typing rather than delayed or choppy responses.
• Supports generating images, embeddings, and configurable languages.
• Built in Deno for better performance, granular permissions, and easier script modification.

Continuing the conversation:

gpt "Give me 5 examples of auto-antonyms"
# Sanction, Bolt, Peruse, Trim, Overlook

gpt -c "Give 5 more"
# Cleave, Dust, Weather, Consult, Off

Regenerating the last assistant message:

gpt "Generate a tweet from the perspective of a cat"
# "Just spent 5 hours napping and now I can't decide if I should eat, stare out the window, or nap some more. #CatLife 🐾😽💤" ...

gpt -r
# Just knocked my hooman's coffee off the counter again... you'd think they'd learn by now nothing is safe at paw level 😹 #LivingLifeOnTheEdge #NineLives

Viewing and resuming past conversation history:

gpt --history
# shellgpt-demo-chat
# cat-tweets
# ...

gpt --history --desc
# ...
# cat-tweets
# shellgpt-demo-chat
# ...

gpt --name "cat-tweets" "Generate more, please"

gpt --dump # Dumps latest entry in the history

Viewing a specific history entry:

gpt --dump --name "2024-04-11_09-52-56"

---

tell me a joke

---

Why couldn't the bicycle stand up by itself? Because it was two-tired!

Setting a system prompt and custom temperature:

gpt --temp 0.85 --max 250 --sys "An AI personal trainer" "What exercises should I do for leg day?"

Increasing the speed of the output using GPT-3.5-turbo and custom WPM:

gpt --fast --wpm 1500 "How can I improve my programming skills?"

Output an image:

gpt --img "a corgi wearing cashmere pants hacking into myspace.com"

Output a vector embedding:

gpt --embed "The quick brown fox jumped over the lazy dog." --dims=1024 --model=text-embedding-3-large

Interactive coding session:

gpt --code --repl "Write a typescript function that prints the first 100 primes"

To develop gpt, first clone the repo, make sure you have the Deno runtime installed (see Installation instructions), and then you can run the code with:

deno run -A mod.ts

For example, if you want to run an interactive chat session, you can do:

deno run -A mod.ts --repl "Write outline for ShellGPT's README.md file"