Terminal-Agent-X

EN / 中文

Features

• 👻 Easy installation and usage with a single command.
• 🎈 Small size, no additional dependencies required.
• 🐼 Supports English and Chinese on Windows CMD, Powershell, Linux shell, etc.
• 🤖 Compatible with GPT-3.5/4, DALL·E, Claude and Gemini APIs or the third parties.

Install

pip install terminal-agent-x

Config

You need to add the following environment variable:

export tax_key=sk-xxx
export tax_base_url=YOUR_BASE_URL # optional, default: https://api.openai.com

Get your key from OpenAI or Claude API key from Anthropic and set it as tax_key.

Get Started

You can use the tax <prompt> to interact with the model, like:

$ tax write a python code for fibonacci

Usage

Model	Command	Description
ChatGPT	tax <prompt>	Use gpt-3.5-turbo to generate content.
ChatGPT	tax <prompt> --code	Use gpt-3.5-turbo to generate code. If only one line is generated, it will be executed automatically with your permission.
GPT-4o-mini	tax <prompt> -m gpt-4o-mini	Use GPT-4o-mini to generate content.
GPT-4o	tax -i image_path -m gpt-4o <prompt>	Upload an image and use GPT-4o to chat.
DALL·E 3	tax <prompt> -m dalle	Use DALL·E 3 to generate image. Currently, only one 1024x1024 image can be generated at a time.
Claude	tax <prompt> -m claude	Use Claude to generate content. Use -k your_claude_key if you have set key in the environment variable. Only support single chat now.
Gemini Pro	tax <prompt> -m gemini-pro	Use Gemini to generate content. Also support -k google_api_key. Only support single chat now.
Gemini Pro (Vision)	tax -i image_path -m gemini-pro-vision <prompt>	Upload an image and use Gemini Vision to chat. Also support -k google_api_key.

Mode	Command	Description
Chat with selected model	tax -m model_name -c	Chat with the selected model.
Use a third-party base url and key	tax <prompt> -u <base_url> -k <key>	The url and key will be use before the environment variable.
Save result to file	tax <prompt> -o <output_file>	Save response to a file.
Read prompts from file	tax --prompt_file <prompt_file>	Read prompt from file, one prompt per line. Please ensure tht you have enough quota.
Parallel processing	tax -p --prompt_file input.txt -o output.txt --option max_workers=3 --option chat_mode=openai	Set more options for parallel processing. For example, --option max_workers=3 means that you can run 3 processes at the same time.

Attention!

You can see a directive after the generated command that says

Do you want to execute the command? (y/n)

Please execute it or not at your own discretion. I am not responsible for the consequences of generated commands.

License

GNU General Public License v3.0

Development Logs

0.1.x

0.1.0

• Implement basic functions
• Support for Windows cmd and Linux shell
• Add --file option for saving the response to a file

0.1.1

• Add --show_all option for showing all contents of the response.
• Add --url option for users not under GFW.
• Add support for Windows Powershell

0.1.2

• Add Anthropic Claude API Support. Thanks to jtsang4/claude-to-chatgpt. (deprecated in 0.1.5)
• Add Support for Chinese on Linux and Windows. (also add a temporary solution for VSCode Terminal on Windows).
• Add a timeout function.
• Fix: C++ code block prefix.

0.1.3

• Fix: code block prefix bug (tax will act maybe a little faster).
• Modify: simplify the code.
• Test: test for multi-process. Now you can use tax more efficiently in terminal.

0.1.4

• Feat: Add support for reading prompt from file.
• Feat: Add support for DALL·E.
• Fix: Resolve the bug of curl command on Windows platform using IPv6 address to access Claude.

0.1.5

• Fix: Change api to a third-party proxy. Affected by GFW's DNS domain pollution, the original proxy is temporarily unavailable. claude-to-chatgpt is unavailable.

0.1.6

• Feat: Add support for Chat on Linux. Now you can use tax as ChatGPT CLI!
• Feat: Add support for native Anthropic Claude API on Linux Shell, Windows cmd and Powershell.

0.1.7

• Feat: Add support for parallel processing with mode.

0.1.8

• Feat: Add support for gpt-4-vision-preview model on all platforms (Beta feature). For example,

  $ tax -i logo.jpg -m gpt-4-vision-preview what is this?
  This appears to be a logo or emblem for something called "Most Creative Learning." The design features a stylized triangular shape, possibly an optical illusion known as a Penrose triangle, ...

• Refactor: Change the way of URL selection and image input.

0.1.9

• Feat: Update DALL·E to dall-e-3 model.
• Refactor: Change the name of environment variable from openai_key to tax_key. And some options are also changed. Please check the help message for more details.
• Fix: Fix the bug of --code option when generating code to file.

0.1.10

• Feat: Add support for Google Gemini Pro for single chat.
• Feat: Add support for Google Gemini Pro Vision (Beta feature).
• Feat: Add support for stream mode of OpenAI.

0.1.11

• Feat: Add support for Anthropic Claude 3 Opus/Sonnet(Vision) API.