jjaimealeman/chat

## readme.md

      
    Raw
  

              readme.md
            
          
    May 02 2023 (date started)

Aug 26 2023 (date modified)

With the explosion of ChatGPT use, I decided to search around for a shell script that I could call from my terminal.
After testing out different ones, I settled on this one: https://github.com/0xacx/chatGPT-shell-cli
It's got some image generation capabilities that I do not use, but I left them there.
I have modified and edited the hell out of it for my own personal use.
The original script saved all previous prompts to a file ~/.chatgpt_history. But I found that after a few weeks of use, that file grew quite large. Mine was over 2,000 lines when I made this revision.
It now creates a new daily file at ~/.chatgpt_history with the format Y-%m-%d.md.
I found this method better as I can now quickly "Live Grep" search with neovim 👍
Filename example: ~/.chat_history/2023-05-02.md.
When prompting the first prompt of the day, this is the output:

Afterwards, this is the output:

Each prompt is now also saved along with the response. I found this format to be easier to read.

All prompt responses piped out to xclip -selection clipboard, so I can quickly and easily paste elsewhere.
Image generation is saved to ~/.chatgpt_history/images folder with date stamp, and suffixed with an incrementing number count.

Usage:

First get your own API key from https://platform.openai.com/account/api-keys
And add these 2 lines to .zshrc.
export OPENAI_API_KEY=sk-**********************************
export PATH="/home/UserName/path/to/script/chat:$PATH"
Give your script the proper permissions.
chmod +x chat
Requirements:

curl (for API calls)
jq (JSON parser)
xclip (clipboard)
glow (markdown viewer)


## chat
#!/bin/bash

GLOBIGNORE="*"

CHAT_INIT_PROMPT="You are ChatGPT, a Large Language Model trained by OpenAI. You will be answering questions from users. You answer as concisely as possible for each response (e.g. don’t be verbose). If you are generating a list, do not have too many items. Keep the number of items short. Before each user prompt you will be given the chat history in Q&A form. Output your answer directly, with no labels in front. Do not start your answers with A or Anwser. You were trained on data up until 2021. Today's date is $(date +%d/%m/%Y)"

SYSTEM_PROMPT="You are ChatGPT, a large language model trained by OpenAI. Answer as concisely as possible. Current date: $(date +%d/%m/%Y). Knowledge cutoff: 9/1/2021."

COMMAND_GENERATION_PROMPT="You are a Command Line Interface expert and your task is to provide functioning shell commands. Return a CLI command and nothing else - do not send it in a code block, quotes, or anything else, just the pure text CONTAINING ONLY THE COMMAND. If possible, return a one-line bash command or chain many commands together. Return ONLY the command ready to run in the terminal. The command should do the following:"

CHATGPT_CYAN_LABEL="\033[36mchatgpt \033[0m"
PROCESSING_LABEL="\n\033[90mProcessing... \033[0m\033[0K\r"
OVERWRITE_PROCESSING_LINE="             \033[0K\r"

if [[ -z "$OPENAI_API_KEY" ]]; then
	echo "You need to set your OPENAI_API_KEY to use this script"
	echo "You can set it temporarily by running this on your terminal: export OPENAI_API_KEY=YOUR_KEY_HERE"
	exit 1
fi

usage() {
	cat <<EOF
A simple, lightweight shell script to use OpenAI's Language Models and DALL-E from the terminal without installing Python or Node.js. Open Source and written in 100% Shell (Bash)

https://github.com/0xacx/chatGPT-shell-cli/

By default the script uses the "gpt-3.5-turbo" model. It will upgrade to "gpt-4" when the API is accessible to anyone.

Commands:
  image: - To generate images, start a prompt with image: If you are using iTerm, you can view the image directly in the terminal. Otherwise the script will ask to open the image in your browser.
  history - To view your chat history
  models - To get a list of the models available at OpenAI API
  model: - To view all the information on a specific model, start a prompt with model: and the model id as it appears in the list of models. For example: "model:text-babbage:001" will get you all the fields for text-babbage:001 model
  command: - To get a command with the specified functionality and run it, just type "command:" and explain what you want to achieve. The script will always ask you if you want to execute the command. i.e.
  "command: show me all files in this directory that have more than 150 lines of code"
  *If a command modifies your file system or dowloads external files the script will show a warning before executing.

Options:
  -i, --init-prompt - Provide initial chat prompt to use in context
  --init-prompt-from-file - Provide initial prompt from file
  -p, --prompt - Provide prompt instead of starting chat
  --prompt-from-file - Provide prompt from file
  -t, --temperature - Temperature
  --max-tokens - Max number of tokens
  -m, --model - Model
  -s, --size - Image size. (The sizes that are accepted by the OpenAI API are 256x256, 512x512, 1024x1024)
  -c, --chat-context - For models that do not support chat context by default (all models except gpt-3.5-turbo and gpt-4), you can enable chat context, for the model to remember your previous questions and its previous answers. It also makes models aware of todays date and what data it was trained on.

EOF
}

# error handling function
# $1 should be the response body
handle_error() {
	if echo "$1" | jq -e '.error' >/dev/null; then
		echo -e "Your request to Open AI API failed: \033[0;31m$(echo $1 | jq -r '.error.type')\033[0m"
		echo $1 | jq -r '.error.message'
		exit 1
	fi
}

# request to OpenAI API completions endpoint function
# $1 should be the request prompt
request_to_completions() {
	request_prompt="$1"

	response=$(curl https://api.openai.com/v1/completions \
		-sS \
		-H 'Content-Type: application/json' \
		-H "Authorization: Bearer $OPENAI_API_KEY" \
		-d '{
  			"model": "'"$MODEL"'",
  			"prompt": "'"${request_prompt}"'",
  			"max_tokens": '$MAX_TOKENS',
  			"temperature": '$TEMPERATURE'
			}')
}

# request to OpenAI API image generations endpoint function
# $1 should be the prompt
request_to_image() {
	prompt="$1"
	image_response=$(curl https://api.openai.com/v1/images/generations \
		-sS \
		-H 'Content-Type: application/json' \
		-H "Authorization: Bearer $OPENAI_API_KEY" \
		-d '{
    		"prompt": "'"${prompt#*image:}"'",
    		"n": 1,
    		"size": "'"$SIZE"'"
			}')
}

# request to OpenAPI API chat completion endpoint function
# $1 should be the message(s) formatted with role and content
request_to_chat() {
	message="$1"
	response=$(curl https://api.openai.com/v1/chat/completions \
		-sS \
		-H 'Content-Type: application/json' \
		-H "Authorization: Bearer $OPENAI_API_KEY" \
		-d '{
            "model": "'"$MODEL"'",
            "messages": [
                {"role": "system", "content": "'"$SYSTEM_PROMPT"'"},
                '"$message"'
                ],
            "max_tokens": '$MAX_TOKENS',
            "temperature": '$TEMPERATURE'
            }')
}

# build chat context before each request for /completions (all models except
# gpt turbo and gpt 4)
# $1 should be the chat context
# $2 should be the escaped prompt
build_chat_context() {
	chat_context="$1"
	escaped_prompt="$2"
	if [ -z "$chat_context" ]; then
		chat_context="$CHAT_INIT_PROMPT\nQ: $escaped_prompt"
	else
		chat_context="$chat_context\nQ: $escaped_prompt"
	fi
	request_prompt="${chat_context//$'\n'/\\n}"
}

# maintain chat context function for /completions (all models except
# gpt turbo and gpt 4)
# builds chat context from response,
# keeps chat context length under max token limit
# $1 should be the chat context
# $2 should be the response data (only the text)
maintain_chat_context() {
	chat_context="$1"
	response_data="$2"
	# add response to chat context as answer
	chat_context="$chat_context${chat_context:+\n}\nA: ${response_data//$'\n'/\\n}"
	# check prompt length, 1 word =~ 1.3 tokens
	# reserving 100 tokens for next user prompt
	while (($(echo "$chat_context" | wc -c) * 1, 3 > (MAX_TOKENS - 100))); do
		# remove first/oldest QnA from prompt
		chat_context=$(echo "$chat_context" | sed -n '/Q:/,$p' | tail -n +2)
		# add init prompt so it is always on top
		chat_context="$CHAT_INIT_PROMPT $chat_context"
	done
}

# build user chat message function for /chat/completions (gpt models)
# builds chat message before request,
# $1 should be the chat message
# $2 should be the escaped prompt
build_user_chat_message() {
	chat_message="$1"
	escaped_prompt="$2"
	if [ -z "$chat_message" ]; then
		chat_message="{\"role\": \"user\", \"content\": \"$escaped_prompt\"}"
	else
		chat_message="$chat_message, {\"role\": \"user\", \"content\": \"$escaped_prompt\"}"
	fi

	request_prompt="$chat_message"
}

# adds the assistant response to the message in (chatml) format
# for /chat/completions (gpt models)
# keeps messages length under max token limit
# $1 should be the chat message
# $2 should be the response data (only the text)
add_assistant_response_to_chat_message() {
	chat_message="$1"
	local local_response_data="$2"

	# replace new line characters from response with space
	local_response_data=$(echo "$local_response_data" | tr '\n' ' ')
	# add response to chat context as answer
	chat_message="$chat_message, {\"role\": \"assistant\", \"content\": \"$local_response_data\"}"

	# transform to json array to parse with jq
	chat_message_json="[ $chat_message ]"
	# check prompt length, 1 word =~ 1.3 tokens
	# reserving 100 tokens for next user prompt
	while (($(echo "$chat_message" | wc -c) * 1, 3 > (MAX_TOKENS - 100))); do
		# remove first/oldest QnA from prompt
		chat_message=$(echo "$chat_message_json" | jq -c '.[2:] | .[] | {role, content}')
	done
}

# parse command line arguments
while [[ "$#" -gt 0 ]]; do
	case $1 in
	-i | --init-prompt)
		CHAT_INIT_PROMPT="$2"
		SYSTEM_PROMPT="$2"
		CONTEXT=true
		shift
		shift
		;;
	--init-prompt-from-file)
		CHAT_INIT_PROMPT=$(cat "$2")
		SYSTEM_PROMPT=$(cat "$2")
		CONTEXT=true
		shift
		shift
		;;
	-p | --prompt)
		prompt="$2"
		shift
		shift
		;;
	--prompt-from-file)
		prompt=$(cat "$2")
		shift
		shift
		;;
	-t | --temperature)
		TEMPERATURE="$2"
		shift
		shift
		;;
	--max-tokens)
		MAX_TOKENS="$2"
		shift
		shift
		;;
	-m | --model)
		MODEL="$2"
		shift
		shift
		;;
	-s | --size)
		SIZE="$2"
		shift
		shift
		;;
	-c | --chat-context)
		CONTEXT=true
		shift
		;;
	-h | --help)
		usage
		exit 0
		;;
	*)
		echo "Unknown parameter: $1"
		exit 1
		;;
	esac
done

# set defaults
TEMPERATURE=${TEMPERATURE:-0.7}
MAX_TOKENS=${MAX_TOKENS:-1024}
MODEL=${MODEL:-gpt-3.5-turbo}
SIZE=${SIZE:-512x512}
CONTEXT=${CONTEXT:-false}

# Create a directory for chat history
chat_history=~/.chat_history/$(date +%Y-%m-%d).md
if [ ! -f "$chat_history" ]; then
	touch "$chat_history"
	chmod 600 "$chat_history"
	echo "Created Chat History: '$chat_history'."
	echo "======================="
else
	prompt_count=$(grep -o "PROMPT" $chat_history | wc -l)
	line_count=$(wc -l <$chat_history)
	echo "Chat history exists: '$chat_history'."
	echo "With $prompt_count prompts and $line_count lines."
	echo "============================="
fi

# Create a directory for chat history images
chat_history_images=~/.chat_history/images
image_count="-type f \( -iname \"*.png\" \) | wc -l"
if [ ! -d "$chat_history_images" ]; then
	mkdir -p "$chat_history_images"
	chmod 755 "$chat_history_images"
	echo "Created Chat History Images Directory: '$chat_history_images'."
	echo "=============================================="
else
	echo "Chat history images directory exists: '$chat_history_images'."
	echo "There are $(find $chat_history_images -type f \( -iname "*.png" \) | wc -l) images."
	echo "=============================================="
fi

running=true
# check input source and determine run mode

# prompt from argument, run on pipe mode (run once, no chat)
if [ -n "$prompt" ]; then
	pipe_mode_prompt=${prompt}
# if input file_descriptor is a terminal, run on chat mode
elif [ -t 0 ]; then
	echo -e "Welcome to chatgpt. You can quit with '\033[36mexit\033[0m' or '\033[36mq\033[0m'."
# prompt from pipe or redirected stdin, run on pipe mode
else
	pipe_mode_prompt+=$(cat -)
fi

while $running; do

	if [ -z "$pipe_mode_prompt" ]; then
		echo -e "\nEnter a prompt:"
		read -e prompt
		if [ "$prompt" != "exit" ] && [ "$prompt" != "q" ]; then
			echo -ne $PROCESSING_LABEL
		fi
	else
		# set vars for pipe mode
		prompt=${pipe_mode_prompt}
		running=false
		CHATGPT_CYAN_LABEL=""
	fi

	if [ "$prompt" == "exit" ] || [ "$prompt" == "q" ]; then
		running=false
	elif [[ "$prompt" =~ ^image: ]]; then
		request_to_image "$prompt"
		handle_error "$image_response"
		image_url=$(echo $image_response | jq -r '.data[0].url')
		echo -e "$OVERWRITE_PROCESSING_LINE"
		echo -e "${CHATGPT_CYAN_LABEL}Your image was created. \n\n"
		echo -e "Link: ${image_url}\n"

		if [[ "$TERM_PROGRAM" == "iTerm.app" ]]; then
			curl -sS $image_url -o temp_image.png
			imgcat temp_image.png
			rm temp_image.png
		elif [[ "$TERM" == "xterm-kitty" ]]; then
			# Get image and name as temp_image.png
			curl -sS $image_url -o temp_image.png
			# Get the number of images in the chat history images directory
			IMAGE_COUNT=$(find $chat_history_images -type f \( -iname "*.png" \) | wc -l)
			echo -e "Image count is: $IMAGE_COUNT\n"
			# Move temp_image.png to chat history images directory with date and image count as part of the file name.
			# EXAMPLE: 2023-08-26_10.png
			mv temp_image.png $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) | wc -l).png
			# Echo path to image.
			echo -e "Local path is: $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) | wc -l).png\n"
			# Display image in kitty terminal.
			kitty +kitten icat $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) | wc -l).png
			# Copy image to clipboard using xclip.
			xclip -selection clipboard -t image/png -i < $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) | wc -l).png
			echo -e "\033[33mImage\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
		else
			echo "Would you like to open it? (Yes/No)"
			read -e answer
			if [ "$answer" == "Yes" ] || [ "$answer" == "yes" ] || [ "$answer" == "y" ] || [ "$answer" == "Y" ] || [ "$answer" == "ok" ]; then
				open "${image_url}"
			fi
		fi
	elif [[ "$prompt" == "history" ]]; then
		# echo -e "\n$(cat ~/.chatgpt_history)"
		echo -e "\n$(cat $chat_history)"
	elif [[ "$prompt" == "models" ]]; then
		models_response=$(curl https://api.openai.com/v1/models \
			-sS \
			-H "Authorization: Bearer $OPENAI_API_KEY")
		handle_error "$models_response"
		models_data=$(echo $models_response | jq -r -C '.data[] | {id, owned_by, created}')
		echo -e "$OVERWRITE_PROCESSING_LINE"
		echo -e "${CHATGPT_CYAN_LABEL}This is a list of models currently available at OpenAI API:\n ${models_data}"
	elif [[ "$prompt" =~ ^model: ]]; then
		models_response=$(curl https://api.openai.com/v1/models \
			-sS \
			-H "Authorization: Bearer $OPENAI_API_KEY")
		handle_error "$models_response"
		model_data=$(echo $models_response | jq -r -C '.data[] | select(.id=="'"${prompt#*model:}"'")')
		echo -e "$OVERWRITE_PROCESSING_LINE"
		echo -e "${CHATGPT_CYAN_LABEL}Complete details for model: ${prompt#*model:}\n ${model_data}"
	elif [[ "$prompt" =~ ^command: ]]; then
		# escape quotation marks
		escaped_prompt=$(echo "$prompt" | sed 's/"/\\"/g')
		# escape new lines
		if [[ "$prompt" =~ ^command: ]]; then
			escaped_prompt=${prompt#command:}
			request_prompt=$COMMAND_GENERATION_PROMPT${escaped_prompt//$'\n'/' '}
		fi
		build_user_chat_message "$chat_message" "$request_prompt"
		request_to_chat "$request_prompt"
		handle_error "$response"
		response_data=$(echo $response | jq -r '.choices[].message.content')

		if [[ "$prompt" =~ ^command: ]]; then
			echo -e "$OVERWRITE_PROCESSING_LINE"
			echo -e "${CHATGPT_CYAN_LABEL} ${response_data}" | fold -s -w $COLUMNS
			dangerous_commands=("rm" ">" "mv" "mkfs" ":(){:|:&};" "dd" "chmod" "wget" "curl")

			for dangerous_command in "${dangerous_commands[@]}"; do
				if [[ "$response_data" == *"$dangerous_command"* ]]; then
					echo "Warning! This command can change your file system or download external scripts & data. Please do not execute code that you don't understand completely."
				fi
			done
			echo "Would you like to execute it? (Yes/No)"
			read run_answer
			if [ "$run_answer" == "Yes" ] || [ "$run_answer" == "yes" ] || [ "$run_answer" == "y" ] || [ "$run_answer" == "Y" ]; then
				echo -e "\nExecuting command: $response_data\n"
				eval $response_data
			fi
		fi
		escaped_response_data=$(echo "$response_data" | sed 's/"/\\"/g')
		add_assistant_response_to_chat_message "$chat_message" "$escaped_response_data"

		# timestamp=$(date +"%d/%m/%Y %H:%M")
		timestamp=$(date +"%Y/%m/%d %H:%M")
		# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
		echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
		echo -e "$response_data" | xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"

	elif [[ "$MODEL" =~ ^gpt- ]]; then
		# escape quotation marks
		escaped_prompt=$(echo "$prompt" | sed 's/"/\\"/g')
		# escape new lines
		request_prompt=${escaped_prompt//$'\n'/' '}

		build_user_chat_message "$chat_message" "$request_prompt"
		request_to_chat "$request_prompt"
		handle_error "$response"
		response_data=$(echo "$response" | jq -r '.choices[].message.content')

		echo -e "$OVERWRITE_PROCESSING_LINE"
		# if glow installed, print parsed markdown
		if command -v glow &>/dev/null; then
			echo -e "${CHATGPT_CYAN_LABEL}"
			echo "${response_data}" | glow -
			#echo -e "${formatted_text}"
		else
			echo -e "${CHATGPT_CYAN_LABEL}${response_data}" | fold -s -w $COLUMNS
		fi
		escaped_response_data=$(echo "$response_data" | sed 's/"/\\"/g')
		add_assistant_response_to_chat_message "$chat_message" "$escaped_response_data"

		# timestamp=$(date +"%d/%m/%Y %H:%M")
		timestamp=$(date +"%Y/%m/%d %H:%M")
		# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
		echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
		echo -e "$response_data" | xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
	else
		# escape quotation marks
		escaped_prompt=$(echo "$prompt" | sed 's/"/\\"/g')
		# escape new lines
		request_prompt=${escaped_prompt//$'\n'/' '}

		if [ "$CONTEXT" = true ]; then
			build_chat_context "$chat_context" "$escaped_prompt"
		fi

		request_to_completions "$request_prompt"
		handle_error "$response"
		response_data=$(echo "$response" | jq -r '.choices[].text')

		echo -e "$OVERWRITE_PROCESSING_LINE"
		# if glow installed, print parsed markdown
		if command -v glow &>/dev/null; then
			echo -e "${CHATGPT_CYAN_LABEL}"
			echo "${response_data}" | glow -
		else
			# else remove empty lines and print
			formatted_text=$(echo "${response_data}" | sed '1,2d; s/^A://g')
			echo -e "${CHATGPT_CYAN_LABEL}${formatted_text}" | fold -s -w $COLUMNS
		fi

		if [ "$CONTEXT" = true ]; then
			escaped_response_data=$(echo "$response_data" | sed 's/"/\\"/g')
			maintain_chat_context "$chat_context" "$escaped_response_data"
		fi

		# timestamp=$(date +"%d/%m/%Y %H:%M")
		timestamp=$(date +"%Y/%m/%d %H:%M")
		# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
		echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
		echo -e "$response_data" | xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
	fi
done
	#!/bin/bash

	GLOBIGNORE="*"

	CHAT_INIT_PROMPT="You are ChatGPT, a Large Language Model trained by OpenAI. You will be answering questions from users. You answer as concisely as possible for each response (e.g. don’t be verbose). If you are generating a list, do not have too many items. Keep the number of items short. Before each user prompt you will be given the chat history in Q&A form. Output your answer directly, with no labels in front. Do not start your answers with A or Anwser. You were trained on data up until 2021. Today's date is $(date +%d/%m/%Y)"

	SYSTEM_PROMPT="You are ChatGPT, a large language model trained by OpenAI. Answer as concisely as possible. Current date: $(date +%d/%m/%Y). Knowledge cutoff: 9/1/2021."

	COMMAND_GENERATION_PROMPT="You are a Command Line Interface expert and your task is to provide functioning shell commands. Return a CLI command and nothing else - do not send it in a code block, quotes, or anything else, just the pure text CONTAINING ONLY THE COMMAND. If possible, return a one-line bash command or chain many commands together. Return ONLY the command ready to run in the terminal. The command should do the following:"

	CHATGPT_CYAN_LABEL="\033[36mchatgpt \033[0m"
	PROCESSING_LABEL="\n\033[90mProcessing... \033[0m\033[0K\r"
	OVERWRITE_PROCESSING_LINE=" \033[0K\r"

	if [[ -z "$OPENAI_API_KEY" ]]; then
	echo "You need to set your OPENAI_API_KEY to use this script"
	echo "You can set it temporarily by running this on your terminal: export OPENAI_API_KEY=YOUR_KEY_HERE"
	exit 1
	fi

	usage() {
	cat <<EOF
	A simple, lightweight shell script to use OpenAI's Language Models and DALL-E from the terminal without installing Python or Node.js. Open Source and written in 100% Shell (Bash)

	https://github.com/0xacx/chatGPT-shell-cli/

	By default the script uses the "gpt-3.5-turbo" model. It will upgrade to "gpt-4" when the API is accessible to anyone.

	Commands:
	image: - To generate images, start a prompt with image: If you are using iTerm, you can view the image directly in the terminal. Otherwise the script will ask to open the image in your browser.
	history - To view your chat history
	models - To get a list of the models available at OpenAI API
	model: - To view all the information on a specific model, start a prompt with model: and the model id as it appears in the list of models. For example: "model:text-babbage:001" will get you all the fields for text-babbage:001 model
	command: - To get a command with the specified functionality and run it, just type "command:" and explain what you want to achieve. The script will always ask you if you want to execute the command. i.e.
	"command: show me all files in this directory that have more than 150 lines of code"
	*If a command modifies your file system or dowloads external files the script will show a warning before executing.

	Options:
	-i, --init-prompt - Provide initial chat prompt to use in context
	--init-prompt-from-file - Provide initial prompt from file
	-p, --prompt - Provide prompt instead of starting chat
	--prompt-from-file - Provide prompt from file
	-t, --temperature - Temperature
	--max-tokens - Max number of tokens
	-m, --model - Model
	-s, --size - Image size. (The sizes that are accepted by the OpenAI API are 256x256, 512x512, 1024x1024)
	-c, --chat-context - For models that do not support chat context by default (all models except gpt-3.5-turbo and gpt-4), you can enable chat context, for the model to remember your previous questions and its previous answers. It also makes models aware of todays date and what data it was trained on.

	EOF
	}

	# error handling function
	# $1 should be the response body
	handle_error() {
	if echo "$1" \| jq -e '.error' >/dev/null; then
	echo -e "Your request to Open AI API failed: \033[0;31m$(echo $1 \| jq -r '.error.type')\033[0m"
	echo $1 \| jq -r '.error.message'
	exit 1
	fi
	}

	# request to OpenAI API completions endpoint function
	# $1 should be the request prompt
	request_to_completions() {
	request_prompt="$1"

	response=$(curl https://api.openai.com/v1/completions \
	-sS \
	-H 'Content-Type: application/json' \
	-H "Authorization: Bearer $OPENAI_API_KEY" \
	-d '{
	"model": "'"$MODEL"'",
	"prompt": "'"${request_prompt}"'",
	"max_tokens": '$MAX_TOKENS',
	"temperature": '$TEMPERATURE'
	}')
	}

	# request to OpenAI API image generations endpoint function
	# $1 should be the prompt
	request_to_image() {
	prompt="$1"
	image_response=$(curl https://api.openai.com/v1/images/generations \
	-sS \
	-H 'Content-Type: application/json' \
	-H "Authorization: Bearer $OPENAI_API_KEY" \
	-d '{
	"prompt": "'"${prompt#*image:}"'",
	"n": 1,
	"size": "'"$SIZE"'"
	}')
	}

	# request to OpenAPI API chat completion endpoint function
	# $1 should be the message(s) formatted with role and content
	request_to_chat() {
	message="$1"
	response=$(curl https://api.openai.com/v1/chat/completions \
	-sS \
	-H 'Content-Type: application/json' \
	-H "Authorization: Bearer $OPENAI_API_KEY" \
	-d '{
	"model": "'"$MODEL"'",
	"messages": [
	{"role": "system", "content": "'"$SYSTEM_PROMPT"'"},
	'"$message"'
	],
	"max_tokens": '$MAX_TOKENS',
	"temperature": '$TEMPERATURE'
	}')
	}

	# build chat context before each request for /completions (all models except
	# gpt turbo and gpt 4)
	# $1 should be the chat context
	# $2 should be the escaped prompt
	build_chat_context() {
	chat_context="$1"
	escaped_prompt="$2"
	if [ -z "$chat_context" ]; then
	chat_context="$CHAT_INIT_PROMPT\nQ: $escaped_prompt"
	else
	chat_context="$chat_context\nQ: $escaped_prompt"
	fi
	request_prompt="${chat_context//$'\n'/\\n}"
	}

	# maintain chat context function for /completions (all models except
	# gpt turbo and gpt 4)
	# builds chat context from response,
	# keeps chat context length under max token limit
	# $1 should be the chat context
	# $2 should be the response data (only the text)
	maintain_chat_context() {
	chat_context="$1"
	response_data="$2"
	# add response to chat context as answer
	chat_context="$chat_context${chat_context:+\n}\nA: ${response_data//$'\n'/\\n}"
	# check prompt length, 1 word =~ 1.3 tokens
	# reserving 100 tokens for next user prompt
	while (($(echo "$chat_context" \| wc -c) * 1, 3 > (MAX_TOKENS - 100))); do
	# remove first/oldest QnA from prompt
	chat_context=$(echo "$chat_context" \| sed -n '/Q:/,$p' \| tail -n +2)
	# add init prompt so it is always on top
	chat_context="$CHAT_INIT_PROMPT $chat_context"
	done
	}

	# build user chat message function for /chat/completions (gpt models)
	# builds chat message before request,
	# $1 should be the chat message
	# $2 should be the escaped prompt
	build_user_chat_message() {
	chat_message="$1"
	escaped_prompt="$2"
	if [ -z "$chat_message" ]; then
	chat_message="{\"role\": \"user\", \"content\": \"$escaped_prompt\"}"
	else
	chat_message="$chat_message, {\"role\": \"user\", \"content\": \"$escaped_prompt\"}"
	fi

	request_prompt="$chat_message"
	}

	# adds the assistant response to the message in (chatml) format
	# for /chat/completions (gpt models)
	# keeps messages length under max token limit
	# $1 should be the chat message
	# $2 should be the response data (only the text)
	add_assistant_response_to_chat_message() {
	chat_message="$1"
	local local_response_data="$2"

	# replace new line characters from response with space
	local_response_data=$(echo "$local_response_data" \| tr '\n' ' ')
	# add response to chat context as answer
	chat_message="$chat_message, {\"role\": \"assistant\", \"content\": \"$local_response_data\"}"

	# transform to json array to parse with jq
	chat_message_json="[ $chat_message ]"
	# check prompt length, 1 word =~ 1.3 tokens
	# reserving 100 tokens for next user prompt
	while (($(echo "$chat_message" \| wc -c) * 1, 3 > (MAX_TOKENS - 100))); do
	# remove first/oldest QnA from prompt
	chat_message=$(echo "$chat_message_json" \| jq -c '.[2:] \| .[] \| {role, content}')
	done
	}

	# parse command line arguments
	while [[ "$#" -gt 0 ]]; do
	case $1 in
	-i \| --init-prompt)
	CHAT_INIT_PROMPT="$2"
	SYSTEM_PROMPT="$2"
	CONTEXT=true
	shift
	shift
	;;
	--init-prompt-from-file)
	CHAT_INIT_PROMPT=$(cat "$2")
	SYSTEM_PROMPT=$(cat "$2")
	CONTEXT=true
	shift
	shift
	;;
	-p \| --prompt)
	prompt="$2"
	shift
	shift
	;;
	--prompt-from-file)
	prompt=$(cat "$2")
	shift
	shift
	;;
	-t \| --temperature)
	TEMPERATURE="$2"
	shift
	shift
	;;
	--max-tokens)
	MAX_TOKENS="$2"
	shift
	shift
	;;
	-m \| --model)
	MODEL="$2"
	shift
	shift
	;;
	-s \| --size)
	SIZE="$2"
	shift
	shift
	;;
	-c \| --chat-context)
	CONTEXT=true
	shift
	;;
	-h \| --help)
	usage
	exit 0
	;;
	*)
	echo "Unknown parameter: $1"
	exit 1
	;;
	esac
	done

	# set defaults
	TEMPERATURE=${TEMPERATURE:-0.7}
	MAX_TOKENS=${MAX_TOKENS:-1024}
	MODEL=${MODEL:-gpt-3.5-turbo}
	SIZE=${SIZE:-512x512}
	CONTEXT=${CONTEXT:-false}

	# Create a directory for chat history
	chat_history=~/.chat_history/$(date +%Y-%m-%d).md
	if [ ! -f "$chat_history" ]; then
	touch "$chat_history"
	chmod 600 "$chat_history"
	echo "Created Chat History: '$chat_history'."
	echo "======================="
	else
	prompt_count=$(grep -o "PROMPT" $chat_history \| wc -l)
	line_count=$(wc -l <$chat_history)
	echo "Chat history exists: '$chat_history'."
	echo "With $prompt_count prompts and $line_count lines."
	echo "============================="
	fi

	# Create a directory for chat history images
	chat_history_images=~/.chat_history/images
	image_count="-type f \( -iname \"*.png\" \) \| wc -l"
	if [ ! -d "$chat_history_images" ]; then
	mkdir -p "$chat_history_images"
	chmod 755 "$chat_history_images"
	echo "Created Chat History Images Directory: '$chat_history_images'."
	echo "=============================================="
	else
	echo "Chat history images directory exists: '$chat_history_images'."
	echo "There are $(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l) images."
	echo "=============================================="
	fi

	running=true
	# check input source and determine run mode

	# prompt from argument, run on pipe mode (run once, no chat)
	if [ -n "$prompt" ]; then
	pipe_mode_prompt=${prompt}
	# if input file_descriptor is a terminal, run on chat mode
	elif [ -t 0 ]; then
	echo -e "Welcome to chatgpt. You can quit with '\033[36mexit\033[0m' or '\033[36mq\033[0m'."
	# prompt from pipe or redirected stdin, run on pipe mode
	else
	pipe_mode_prompt+=$(cat -)
	fi

	while $running; do

	if [ -z "$pipe_mode_prompt" ]; then
	echo -e "\nEnter a prompt:"
	read -e prompt
	if [ "$prompt" != "exit" ] && [ "$prompt" != "q" ]; then
	echo -ne $PROCESSING_LABEL
	fi
	else
	# set vars for pipe mode
	prompt=${pipe_mode_prompt}
	running=false
	CHATGPT_CYAN_LABEL=""
	fi

	if [ "$prompt" == "exit" ] \|\| [ "$prompt" == "q" ]; then
	running=false
	elif [[ "$prompt" =~ ^image: ]]; then
	request_to_image "$prompt"
	handle_error "$image_response"
	image_url=$(echo $image_response \| jq -r '.data[0].url')
	echo -e "$OVERWRITE_PROCESSING_LINE"
	echo -e "${CHATGPT_CYAN_LABEL}Your image was created. \n\n"
	echo -e "Link: ${image_url}\n"

	if [[ "$TERM_PROGRAM" == "iTerm.app" ]]; then
	curl -sS $image_url -o temp_image.png
	imgcat temp_image.png
	rm temp_image.png
	elif [[ "$TERM" == "xterm-kitty" ]]; then
	# Get image and name as temp_image.png
	curl -sS $image_url -o temp_image.png
	# Get the number of images in the chat history images directory
	IMAGE_COUNT=$(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l)
	echo -e "Image count is: $IMAGE_COUNT\n"
	# Move temp_image.png to chat history images directory with date and image count as part of the file name.
	# EXAMPLE: 2023-08-26_10.png
	mv temp_image.png $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l).png
	# Echo path to image.
	echo -e "Local path is: $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l).png\n"
	# Display image in kitty terminal.
	kitty +kitten icat $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l).png
	# Copy image to clipboard using xclip.
	xclip -selection clipboard -t image/png -i < $chat_history_images/$(date +%Y-%m-%d)_$(find $chat_history_images -type f \( -iname "*.png" \) \| wc -l).png
	echo -e "\033[33mImage\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
	else
	echo "Would you like to open it? (Yes/No)"
	read -e answer
	if [ "$answer" == "Yes" ] \|\| [ "$answer" == "yes" ] \|\| [ "$answer" == "y" ] \|\| [ "$answer" == "Y" ] \|\| [ "$answer" == "ok" ]; then
	open "${image_url}"
	fi
	fi
	elif [[ "$prompt" == "history" ]]; then
	# echo -e "\n$(cat ~/.chatgpt_history)"
	echo -e "\n$(cat $chat_history)"
	elif [[ "$prompt" == "models" ]]; then
	models_response=$(curl https://api.openai.com/v1/models \
	-sS \
	-H "Authorization: Bearer $OPENAI_API_KEY")
	handle_error "$models_response"
	models_data=$(echo $models_response \| jq -r -C '.data[] \| {id, owned_by, created}')
	echo -e "$OVERWRITE_PROCESSING_LINE"
	echo -e "${CHATGPT_CYAN_LABEL}This is a list of models currently available at OpenAI API:\n ${models_data}"
	elif [[ "$prompt" =~ ^model: ]]; then
	models_response=$(curl https://api.openai.com/v1/models \
	-sS \
	-H "Authorization: Bearer $OPENAI_API_KEY")
	handle_error "$models_response"
	model_data=$(echo $models_response \| jq -r -C '.data[] \| select(.id=="'"${prompt#*model:}"'")')
	echo -e "$OVERWRITE_PROCESSING_LINE"
	echo -e "${CHATGPT_CYAN_LABEL}Complete details for model: ${prompt#*model:}\n ${model_data}"
	elif [[ "$prompt" =~ ^command: ]]; then
	# escape quotation marks
	escaped_prompt=$(echo "$prompt" \| sed 's/"/\\"/g')
	# escape new lines
	if [[ "$prompt" =~ ^command: ]]; then
	escaped_prompt=${prompt#command:}
	request_prompt=$COMMAND_GENERATION_PROMPT${escaped_prompt//$'\n'/' '}
	fi
	build_user_chat_message "$chat_message" "$request_prompt"
	request_to_chat "$request_prompt"
	handle_error "$response"
	response_data=$(echo $response \| jq -r '.choices[].message.content')

	if [[ "$prompt" =~ ^command: ]]; then
	echo -e "$OVERWRITE_PROCESSING_LINE"
	echo -e "${CHATGPT_CYAN_LABEL} ${response_data}" \| fold -s -w $COLUMNS
	dangerous_commands=("rm" ">" "mv" "mkfs" ":(){:\|:&};" "dd" "chmod" "wget" "curl")

	for dangerous_command in "${dangerous_commands[@]}"; do
	if [[ "$response_data" == "$dangerous_command" ]]; then
	echo "Warning! This command can change your file system or download external scripts & data. Please do not execute code that you don't understand completely."
	fi
	done
	echo "Would you like to execute it? (Yes/No)"
	read run_answer
	if [ "$run_answer" == "Yes" ] \|\| [ "$run_answer" == "yes" ] \|\| [ "$run_answer" == "y" ] \|\| [ "$run_answer" == "Y" ]; then
	echo -e "\nExecuting command: $response_data\n"
	eval $response_data
	fi
	fi
	escaped_response_data=$(echo "$response_data" \| sed 's/"/\\"/g')
	add_assistant_response_to_chat_message "$chat_message" "$escaped_response_data"

	# timestamp=$(date +"%d/%m/%Y %H:%M")
	timestamp=$(date +"%Y/%m/%d %H:%M")
	# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
	echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
	echo -e "$response_data" \| xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"

	elif [[ "$MODEL" =~ ^gpt- ]]; then
	# escape quotation marks
	escaped_prompt=$(echo "$prompt" \| sed 's/"/\\"/g')
	# escape new lines
	request_prompt=${escaped_prompt//$'\n'/' '}

	build_user_chat_message "$chat_message" "$request_prompt"
	request_to_chat "$request_prompt"
	handle_error "$response"
	response_data=$(echo "$response" \| jq -r '.choices[].message.content')

	echo -e "$OVERWRITE_PROCESSING_LINE"
	# if glow installed, print parsed markdown
	if command -v glow &>/dev/null; then
	echo -e "${CHATGPT_CYAN_LABEL}"
	echo "${response_data}" \| glow -
	#echo -e "${formatted_text}"
	else
	echo -e "${CHATGPT_CYAN_LABEL}${response_data}" \| fold -s -w $COLUMNS
	fi
	escaped_response_data=$(echo "$response_data" \| sed 's/"/\\"/g')
	add_assistant_response_to_chat_message "$chat_message" "$escaped_response_data"

	# timestamp=$(date +"%d/%m/%Y %H:%M")
	timestamp=$(date +"%Y/%m/%d %H:%M")
	# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
	echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
	echo -e "$response_data" \| xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
	else
	# escape quotation marks
	escaped_prompt=$(echo "$prompt" \| sed 's/"/\\"/g')
	# escape new lines
	request_prompt=${escaped_prompt//$'\n'/' '}

	if [ "$CONTEXT" = true ]; then
	build_chat_context "$chat_context" "$escaped_prompt"
	fi

	request_to_completions "$request_prompt"
	handle_error "$response"
	response_data=$(echo "$response" \| jq -r '.choices[].text')

	echo -e "$OVERWRITE_PROCESSING_LINE"
	# if glow installed, print parsed markdown
	if command -v glow &>/dev/null; then
	echo -e "${CHATGPT_CYAN_LABEL}"
	echo "${response_data}" \| glow -
	else
	# else remove empty lines and print
	formatted_text=$(echo "${response_data}" \| sed '1,2d; s/^A://g')
	echo -e "${CHATGPT_CYAN_LABEL}${formatted_text}" \| fold -s -w $COLUMNS
	fi

	if [ "$CONTEXT" = true ]; then
	escaped_response_data=$(echo "$response_data" \| sed 's/"/\\"/g')
	maintain_chat_context "$chat_context" "$escaped_response_data"
	fi

	# timestamp=$(date +"%d/%m/%Y %H:%M")
	timestamp=$(date +"%Y/%m/%d %H:%M")
	# echo -e "$timestamp \n$prompt \n$response_data \n" >>~/.chatgpt_history
	echo -e "================\n$timestamp \n----------------\nPROMPT:\n$prompt \n\nRESPONSE:\n$response_data \n" >>$chat_history
	echo -e "$response_data" \| xclip -selection clipboard
	echo -e "\033[33mResponse\033[0m \033[33mon\033[0m \033[33mclipboard\033[0m 👍"
	fi
	done