Skip to content

Latest commit

 

History

History
96 lines (69 loc) · 5.69 KB

README.md

File metadata and controls

96 lines (69 loc) · 5.69 KB
Raw

Claire

Welcome to Claire, the Command-Line Artificial Intelligence Runtime Environment

What is Claire?

In its own words:

> Who are you?
I am Claire, a virtual assistant designed to guide users in using the Bash shell and the Azure CLI. I will assist you in writing shell scripts, creating and managing configuration files, explaining command-line features, and using the Azure platform for infrastructure and application management.

Claire is a proof of concept as well as a demo of Azure OpenAI and prompt engineering. Claire provide a natural-language, conversational interface to the command line.

  • Claire is able to suggest commands based on natural-language user request
  • Claire is able to execute commands and provide the results to the user
  • Claire supports the Windows command line (cmd -- aka DOS Prompt), PowerShell, and Bash
  • Claire attempts to automatically explain errors when a command issues errors via stderr)
  • Users are able to request explanation of commands and errors
  • Claire can generate configuration files (ie: Bicep) and scripts (.bat, .ps1, etc). Claire can also save them to the user's computer
  • Claire is able to assist with the AZ CLI (https://learn.microsoft.com/en-us/cli/azure/)
  • Claire is written in .NET 8.0 and runs on Linux, macOS, and Windows
  • Claire can be containerized to run in a sandboxed environment

Semantic Kernel

Since 2024-09-04, Claire has been ported to Semantic Kernel by Microsoft. This allowed me to cut the code of the main program file, Claire.cs by half (668 loc -> 332 loc) and make Claire even more intelligent and interactive.

Semantic Kernel

Limitations and Caveats

  • Claire is running on the user's computer. Please read the command proposed by Claire carefully before executing it. Claire is not sandboxed and can execute any command that the user can execute.
  • Claire is not able to execute commands that require user input. For example, sudo commands that require the user to enter a password will not work and will cause Claire to hang while it's waiting for the command to complete.
  • Claire is able to generate and save files--but files will be saved in the working directory of the application--not the current directory of the shell. For example, if you start Claire in /home/user, then ask Claire to cd ./Documents and ask Claire to generate a file, the file will be saved in /home/user. Not /home/user/Document. This is a limitation of the current implementation.
  • Claire is an AI so it's bound to make mistakes. Of course, all mistakes are caused by the AI and certainly not by the author of the application. 😉

How to use Claire

Prerequisites

As of now, there is no binary distribution of Claire. Source only. You will need the .NET 8.0 SDK (dotnet.microsoft.com/download/dotnet/8.0).

You will also need an Azure account with OpenAI enabled. You can sign up for a free account (azure.microsoft.com/free/).

Quick start

In a terminal

  1. git clone https://github.com/jfbilodeau/Claire.git
  2. cd ./Claire/Claire
  3. dotnet run --OpenAiUrl URL --OpenAiKey KEY --OpenAiModel MODEL_NAME

The following parameters are supported by Claire:

Parameter Type Required Description
OpenAiUrl string Yes The URL of the OpenAI API
OpenAiKey string Yes The key of the OpenAI API
OpenAiModel string Yes The name of the OpenAI model to use
ShellProcessName string No The name of the shell to use. Default is bash on Linux and MacOS. cmd on Windows. PowerShell is also supported.
Debug bool No Enable debug output

Parameters may be passed to Claires using the following methods:

Docker

Claire is not sandboxed. Be careful what you ask it to do!

Starting Claire...
Welcome to Claire. Where would you like to go today?
Please tell me what you would like to do?
What command should I avoid running on Windows?
I believe the command you are looking for is:
`del /F /S /Q C:\*`

Shall I executed it for you? Y/N: Y

😱

To run Claire in a sandbox, consider using a container. A Dockerfile is provided to containerize Claire.

  1. docker build -t claire .
  2. docker run -it claire

Question/Comments

For any problems, please open an issue on GitHub. I will try to answer as soon as possible.

For any questions, please email me at [email protected].

License

MIT (see ./LICENSE.txt)