Chat with GPT in Terminal

This project enables chatting with ChatGPT in the terminal.

Markdown content in answers is rendered as beautifully formatted rich text.

Supports history retrieval with the up arrow key, optional multi-line questions, and tokens counting.

Slash (/) commands are available in the chat box to toggle multi-line submit mode, undo the last question and answer, modify the system prompt and more, see the available commands below for details.

Supports saving chat messages to a JSON file and loading them from the file.

Uses the gpt-3.5-turbo model, which is the same model used by ChatGPT (Free Edition), as default.

Related Projects

GPTerm implemented in C/C++ by @Ace-Radom

gpt-term that can call POE API by @Lemon-2333

Changelog

2023-11-17

• Added the feature of setting the default model. Now you can use gpt-term --set-model MODEL_NAME to set the default model. This setting will be written into the configuration file and will take effect every time it is started.

• Added new model candidates: gpt-4-1106-preview, gpt-4-vision-preview, gpt-3.5-turbo-1106.

2023-06-21

• Added direct query mode, now you can run gpt-term with the content of the question as a parameter to conduct a single question and answer

  gpt-term "What's the weather like today?"

  The answer will be printed directly, or piped into a variable

  gpt-term "What's the weather like today" | read answer
  echo $answer

More Change log

2023-05-20

• Added host configuration support, which is very useful when using self-built API reverse proxy server (#49), you can now use gpt-term --set-host HOST to configure host, the default is https://api.openai.com

2023-05-18

• Added multi-language support: English, Chinese, Japanese, German, follow the system language by default, now you can use /lang to switch languages

2023-05-11

• Find the command the user most likely intended to enter when typing an unrecognized command

2023-05-05

• Add /rand command to set temperature parameter

• Add overflow mode switch for /stream command, now you can run command /stream visible to switch to always visible mode. In this mode, the content that exceeds the screen will be scrolled up, and the new content will be output until it is completed

2023-04-23

Released gpt-term on Pypi, started version control. No need to clone the project locally anymore, simply use the pip command to install gpt-term.

2023-04-15

• Added the ability to create a line break in single-line mode using Esc + Enter

2023-04-13

• Added the function of generating and setting the terminal title in the background, and now the client will use the summary of the first question content as the terminal title

2023-04-09

• Add filename generate function, client will suggest the summary of the first question as filename when save command executed.

2023-04-05

• Add /delete command to delete the first question and answer in this chat to reduce token.

2023-04-01

• Add /copy command to copy the last reply's content to the clipboard
• Add streaming output mode, enabled by default, use /stream to switch

2023-03-28

• Add --model runtime argument and /model command to choose / change AI models.

2023-03-27

• Added --key runtime argument to select which API key in the .env file to use.

2023-03-23

• Added slash (/) command functionality
• Added --load runtime argument to load previously saved chat history
• Modified program structure and interaction methods, changing the original input() function to the prompt_toolkit library's input interface, supporting multi-line input, command-line completion, and other features.
• Improved error handling mechanisms, added chat history backup, logging, and other features, enhancing the program's reliability and fault tolerance.
• Refactored code logic and function structure, improving modularity and readability.

Preparation

1. An OpenAI API key. You need to register an OpenAI account and obtain an API key.

   OpenAI's API key can be generated on the page opened by clicking "View API keys" in the upper right corner of the homepage, direct link: https://platform.openai.com/account/api-keys

   

2. Python version 3.7 or higher.

   Attention: Try not to use the Python that comes with the system (including Windows 11 app store version and MacOS pre-installed Python), otherwise the gpt-term command will not be found after installation (#38)

Installation

1. Install GPT-Term using pip

   pip3 install gpt-term

2. Configure the API Key

   gpt-term --set-apikey YOUR_API_KEY

   If you don't configure the API Key now, you can enter it when prompted during runtime.

Update

To update GPT-Term to the latest version, run the following command in your terminal:

pip3 install --upgrade gpt-term

If there is a new version, GPT-Term will prompt the user to update upon exiting.

How to Use

Run with the following command:

gpt-term

Or:

python3 -m gpt_term

Quick query:

gpt-term "What's the weather like today?"

When entering a question in single-line mode, use Esc + Enter to start a new line, and use Enter to submit the question.

Here are some common shortcut keys (also shortcut keys for the shell):

• Ctrl+_: Undo
• Ctrl+L: Clear screen, equivalent to clear command in shell
• Ctrl+C: Stop the current answer or cancel the current input
• Tab: Autocomplete commands or parameters
• Ctrl+U: Delete all characters to the left of the cursor
• Ctrl+K: Delete all characters to the right of the cursor
• Ctrl+W: Delete the word to the left of the cursor

Original chat logs will be saved to ~/.gpt-term/chat.log

Available Arguments

Arguments	Description	Examples
-h, --help	show this help message and exit	gpt-term --help
--load FILE	Load chat history from file	gpt-term --load chat_history_code_check.json
--key API_KEY	Select the API key to use in the config.ini file	gpt-term --key OPENAI_API_KEY1
--model MODEL	Select AI model to use	gpt-term --model gpt-3.5-turbo
--host HOST	Set the API Host address used in this run (this is usually used to configure proxy)	gpt-term --host https://closeai.deno.dev
-m, --multi	Enable multiline mode	gpt-term --multi
-r, --raw	Enable raw mode	gpt-term --raw
-l, --lang LANG	Set the current running language: en, zh_CN, jp, de	gpt-term --lang en
--set-model HOST	Set the AI model to use	gpt-term --set-model gpt-4-1106-preview
--set-host HOST	Set API Host address (this is usually used to configure proxy)	gpt-term --set-host https://closeai.deno.dev
--set-apikey KEY	Set OpenAI API key	gpt-term --set-apikey sk-xxx
--set-timeout SEC	Set the maximum wait time for API requests	gpt-term --set-timeout 10
--set-gentitle BOOL	Set whether to auto-generate title for chat	gpt-term --set-gentitle True
--set-saveperfix PERFIX	Set the save prefix for chat history files	gpt-term --set-saveperfix chat_history_
--set-loglevel LEVEL	Set log level: DEBUG, INFO, WARNING, ERROR, CRITICAL	gpt-term --set-loglevel DEBUG
--set-lang LANG	Set language: en, zh_CN, jp, de	gpt-term --set-lang en

Multi-line mode and raw mode can be used simultaneously

Configuration File

The configuration file is located at ~/.gpt-term/config.ini and is autogenerated. It can be modified using the program's --set option or edited manually.

The default configuration is as follows:

[DEFAULT]
# API key for OpenAI
OPENAI_API_KEY=

# The maximum waiting time for API requests, the default is 30s
OPENAI_API_TIMEOUT=30

# Whether to automatically generate titles for conversations, enabled by default (generating titles will consume a small amount of tokens)
AUTO_GENERATE_TITLE=True

# Define the default file prefix when the /save command saves the chat history. The default value is "./chat_history_", which means that the chat history will be saved in the file starting with "chat_history_" in the current directory
# At the same time, the prefix can also be specified as a directory + / to allow the program to save the chat history in a folder (note that the corresponding folder needs to be created in advance), for example: CHAT_SAVE_PERFIX=chat_history/
CHAT_SAVE_PERFIX=./chat_history_

# Log level, default is INFO, available value: DEBUG, INFO, WARNING, ERROR, CRITICAL
LOG_LEVEL=INFO

# Set the language of the program, the default is empty, it will follow the system language
LANGUAGE=

# 设置api的host，默认为https://api.openai.com
OPENAI_HOST=

# 设置使用的模型，默认为gpt-3.5-turbo
# 可用模型: https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo, https://platform.openai.com/docs/models/gpt-3-5
OPENAI_MODEL=

Available Commands

• /raw: Display raw text in replies instead of rendered Markdown format

  After switching, use the /last command to reprint the last reply

• /multi: Enable or disable multi-line mode, allowing users to enter multi-line text

  In multi-line mode, use Esc + Enter to submit the question

  If pasting multi line text, single-line mode can also paste properly

• /stream: disable or enable stream mode

  In stream mode, the answer will start outputting as soon as the first response arrives, which can reducing waiting time. Stream mode is on by default.

  • /stream ellipsis (default)

    Switch the mode of streaming output to auto omission, when the output content exceeds the screen, three small dots will be displayed at the bottom of the screen and wait until the output is completed

  • /stream visible

    Toggle the streaming output mode to always visible, in this mode, the content that exceeds the screen will be scrolled up, and the new content will be output until it is completed. Note that in this mode the terminal will not properly clean up off-screen content.

• /tokens: Display the total tokens spent and the tokens for the current conversation

  GPT-3.5 has a token limit of 4096; use this command to check if you're approaching the limit

• /usage: Display the API credits summary

  This feature may not be stable. If it fails to operate, you can visit the usage page to view further information.

• /model: Show or change the Model in use

  gpt-4 , gpt-4-32k , gpt-3.5-turbo are supported by default. when using other models you need to change the API endpoint in code.

• /last: Show the last reply

• /copy or /copy all: Copy the last reply's content to the clipboard

  • /copy code [index]: Copy the index-th code block from the last reply's content to the clipboard

    If index is not specified, the terminal will print all code blocks and ask for the number of the one to be copied

• /delete or /delete first: delete the first question and answer in the current chat

  When the token is about to reach the upper limit, the user will be warned, and when the upper limit has been exceeded, it will be asked whether to delete the first message

  • /delete all: delete all messages

• /save [filename_or_path]: Save the chat history to the specified JSON file

  If no filename or path is provided, the client will generate one, and if generation fails, the filename chat_history_YEAR-MONTH-DAY_HOUR,MINUTE,SECOND.json is suggested on input.

• /system [new_prompt]: Modify the system prompt

• /rand [randomness]: Set Model sampling temperature (0~2), higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.

• /title [new_title]: Set terminal title for this chat

  If new_title is not provided, a new title will be generated based on first question

• /timeout [new_timeout]: Modify API timeout.

  The default timeout is 30 seconds, it can also be configured by setting OPENAI_API_TIMEOUT= in the ~/.gpt-term/config.ini file.

• /undo: Delete the previous question and answer

• /version: Display the local and remote versions of GPT-Term

• /help: Display available commands

• /exit: Exit the application

Exit Words

In the chat, use exit words to end the current session. Exit words include:

['再见', 'bye', 'goodbye', '结束', 'end', '退出', 'exit', 'quit']

Exit words will be sent as a question to ChatGPT, and the application will exit after GPT replies.

You can also use Ctrl-D or /exit to exit immediately.

Upon exit, the token count for the chat session will be displayed.

Current price: $0.002 / 1K tokens, Free Edition rate limit: 20 requests / min (gpt-3.5-turbo)

Dependencies

Thanks to the following projects for providing strong support for this script:

• requests: For handling HTTP requests
• pyperclip: A cross-platform clipboard operation library
• rich: For outputting rich text in the terminal
• prompt_toolkit: Command-line input processing library
• sseclient-py: For implementing streaming output of answers
• tiktoken: A library for calculating and processing OpenAI API tokens

Contributing

Feel free to dive in! Open an issue or submit PRs.

Contributors

This project exists thanks to all the people who contribute.

Project Structure

.
├── LICENSE                   # License
├── README.md                 # Documentation
├── chat.py                   # Script entry point
├── gpt_term                  # Project package folder
│   ├── __init__.py
│   ├── config.ini            # API key storage and other settings
│   └── main.py               # Main program
├── requirements.txt          # List of dependencies
└── setup.py

License

This project is licensed under the MIT License.