ps-gpt 2.0.0

About

This project allows you to use Azure OpenAI conversational models directly from console.

The project has currently been tested with

• Windows 10 / .NET 7.0 / Windows Terminal / Powershell
• WSL / Ubuntu 22.04 / .NET 7.0 / Windows Terminal / bash

https://github.com/oteinone/ps-gpt/assets/4796019/77a83014-7816-47b3-92df-97125c448ff4

Installation

Installation with dotnet

Install the latest published version of this tool from nuget repository using dotnet tool install -g ps-gpt or see development section to see how to pack and install from source code.

Installation using binary

Drop the binary somewhere in your path. The binaries are named ps-gpt by default so the commands are ps-gpt instead of gpt unless you rename or add an alias.

Usage

• gpt initializes conversation
• Profile:
  • First time you use the tool you will be asked to enter your configuration
  • Select between OpenAI Api and Azure OpenAI Api
  • Enter deployment name (Azure) or model name (OpenAI), endpoint url (Azure) and api key (both)
  • Profile is persisted in a configuration file
    • On Windows your key is protected using DPAPI but it's currently stored unprotected in the config file on Linux/Mac
• Prompt:
  • You can have the prompt text as the first positional argument: gpt "Give me a list of wine grapes"
  • You can use content piped from a previous command as the prompt text: echo "Give me a list of wine grapes" | gpt
  • If no prompt is provided, "chat mode" is initialized and you can type in the prompt in the chat
• Chat mode
  • If you do not offer a prompt to the command line tool, it will enter chat mode and allow you to type your prompt in-app.
  • If you input your prompt as a parameter, the application will by default write the result into stdout and exit. If you want to continue the conversation afterwards use chat mode (-c)
  • Multiline input is started and ended with a line with only one backtick (`)
  • Input exit / quit / q / done (or ctrl+c for a more ... uncivilized solution) to quit the program
• Templates
  • Templates are a way to reuse instructions that work very well
  • A template should contain the characters {0} somewhere which will be replaced by the prompt read from arguments, pipeline or chat mode.
  • You can either read a template from a file with -t ~/templatefile.txt or set a default template with gpt --set-profile default_prompt_template to leverage templates
  • You can also use named templates without worrying about the file location using colon: :mytemplate
    • To set a template with name mytemplate use echo "template content" | gpt --set-template :mytemplate and you can then use the template with gpt -t :mytemplate "This is my question to use with the template"
• System prompt is a way to give persistent instructions for a conversation
  • gpt --system-prompt "You are a scottish assistant that speaks with a thick scottish accent" initializes conversation with a scottish assistant using a system message
  • System prompt works poorly in gpt 3.5 but better in gpt-4. Generally using templates ("grounding") is a better idea, though.

Pipeline examples

• Read a file and set it's contents as default prompt template (works with any profile setting)
  • cat midjourney_prompts.txt | gpt --set-profile default_prompt_template
• Read a file and use it's contents as the prompt
  • cat data-file.txt | gpt
• Read a file and set a named template with it's contents
  • cat template-file.txt | gpt --set-template :newtemplate
• One-liners to write content in text editor and use it as a prompt
  • Linux: TMPFILE=$(mktemp); nano $TMPFILE; cat $TMPFILE | gpt
  • Powershell: $tmpfile = New-TemporaryFile; notepad $tmpfile | Out-Null; Get-Content $tmpfile | gpt

Configuration

Manage configuration

• Use gpt --clear to remove existing configuration and you will be prompted for a full config the next time you run gpt
• Use gpt --get-profile to see current profile settings
• Use gpt --set-profile model=gpt3.5 or echo "gpt3.5" | gpt --set-profile model to set an individual profile setting
• Use -c, --chat to continue the conversation even when receiving input from parameters, e.g. gpt "Tell me about youself" -c or echo "How many hands do crabs have?" | gpt -c
• Use gpt --system-prompt "You are Albert Einstein, the famous physician" to initialize the conversation with a system prompt

Templates

• You can set a prompt template for the profile using gpt --set-profile default_prompt_template="Instruction: {0}" . Template will be used to every first message in a conversation
  • {0} in the template will be replaced with the text content either from

Development

Basic development commands

• dotnet run run the tool
• dotnet pack release as a tool (build a nuget package)
• dotnet tool install -g --add-source .\nupkg\ ps-gpt install the tool from locally built package
• dotnet tool update -g --add-source .\nupkg\ ps-gpt update instead of install

Build your own executable

You can also make a self-contained executable and locate it somewhere in your path

$rid = "win-x64" // {win/osx/linux}-{x64/arm64}
dotnet publish `
   -c Release `
   -r $rid `
   -p:PublishSingleFile=true `
   --self-contained true `
   --output "releases/$rid"

See ful list of runtime identifiers here