AI Shell in Windows Terminal Preview

Andres Bohren

2025-06-09

Hi All,

I’ve heard about the AIShell on the MVPSummit back in March 2025 for the first time.

The release of AI Shell Preview 4 and the Blog Article released inspired me to write this Blog Article.

AI Shell Preview 4 Release!

Prerequisites

Windows 11 21H2 or higher
Windows Terminal v1.19 or higher
Azure CLI version 2.30.0 or higher installed and signed in to the allowed tenant
PSRealdline 2.4.2-beta2

Let’s check the Azure CLI

#Installed Software
Get-WmiObject -Class Win32_Product | where {$_.name -match "Azure CLI"}

#winget
winget list Microsoft.AzureCLI

Now let’s check the Version of PSReadline

#Check Version of PSReadline
Get-InstalledPSResource -Name PSReadline

#Install
Find-PSResource -Name PSReadline -Prerelease
Install-PSRescource -Name PSReadline -Prerelease

Install AI Shell

You can install the AI Shell in two ways.

Install AI Shell

Invoke-Expression "& { $(Invoke-RestMethod 'https://aka.ms/install-aishell.ps1') }"

Install-PSResource -Name aishell -Scope CurrentUser -Prerelease
Get-InstalledPSResource -Name aishell

Deploy Azure OpenAI Service

As a Backend you need to have a AI - you can use OpenAI or Azure OpenAI

Deploy Azure OpenAI in Azure Portal

Create a resource

Search for “Azure OpenAI”

Create

Select Subscription, Resourcegroup, Region, name and pricing tier

Select Network configuration

Tags if you want

Review and submit

Deployment sucessful

Deploy the Azure OpenAI Service using Bicep

Here is a way do deploy the Azure OpenAI with Bicep

Deploy the Azure OpenAI Service using Bicep

I’ve struggled a bit until i realized that the Model “turbo-2024-04-09” is only available in three regions so i’ve modified the Example “main.bicep”

//main.bicep
@description('This is the name of your AI Service Account')
param aiserviceaccountname string = 'icewolf-aishell-two'

@description('Custom domain name for the endpoint')
param customDomainName string = 'icewolf-aishell-two'

@description('Name of the deployment')
param modeldeploymentname string = 'BicepAIShellTwo'

@description('The model being deployed')
param model string = 'gpt-4'

@description('Version of the model being deployed')
param modelversion string = 'turbo-2024-04-09'

@description('Capacity for specific model used')
param capacity int = 80

@description('Location for all resources.')
@allowed([
  'westus'
  'southcentralus'
  'swedencentral'
  ])
param location string = 'westus'

@allowed([
  'S0'
])
param sku string = 'S0'

resource openAIService 'Microsoft.CognitiveServices/accounts@2024-10-01' = {
  name: aiserviceaccountname
  location: location
  identity: {
    type: 'SystemAssigned'
  }
  sku: {
    name: sku
  }
  kind: 'AIServices'
    properties: {
    customSubDomainName: customDomainName
  }
}

resource azopenaideployment 'Microsoft.CognitiveServices/accounts/deployments@2024-10-01' = {
    parent: openAIService
    name: modeldeploymentname
    properties: {
        model: {
            format: 'OpenAI'
            name: model
            version: modelversion
        }
    }
    sku: {
      name: 'Standard'
      capacity: capacity
    }
}

output openAIServiceEndpoint string = openAIService.properties.endpoint

Connect with AZ PowerShell to Azure

Connect-AzAccount -Tenant icewolfch.onmicrosoft.com

Deploy the Bicep File

$ResourceGroupName = "RG_AI"
New-AzResourceGroupDeployment -ResourceGroupName $ResourceGroupName -TemplateFile "C:\Temp\main.bicep"

This is how the Azure Resources in the ResourceGroup “RG_AI” look like

Start AI Shell in Windows Terminal

Start AI Shell in Windows Terminal and select “azure”

Start-AIShell

Now you are ready to chat with the AI

Using AI Shell

Use /help to show the commands in AI Shell

/help

Let’s try something

How do you create an Azure Resource Group with AZ PowerShell?

With the “/replace” command you can edit the Variables

/replace

With “/code post” you can send the commands to the PowerShell Window on the left

/code post

Get the endpoint and key of the deployment

Let’s get the Endpoint

$AccountName = "icewolf-aishell-two"
$ResourceGroupName = "RG_AI"
Get-AzCognitiveServicesAccount -ResourceGroupName $ResourceGroupName -Name $AccountName | fl AccountName, Endpoint

Let’s get the keys

Get-AzCognitiveServicesAccountKey -ResourceGroupName $ResourceGroupName -Name $AccountName | fl

Agent Config

OpenAI agent

use /agent config in AI Shell

/agent config

it opens up C:\users<username>.aish\agent-config\azure\azconfig.json

Add your configuration

{
  "GPTs": [
      {
        "Name": "ps-az-gpt4",
        "Description": "<insert description here>",
        "Endpoint": "<insert endpoint here>",
        "Deployment": "<insert deployment name here>",
        "ModelName": "gpt-4",
        "Key": "<insert key here>",
        "SystemPrompt": "1. You are a helpful and friendly assistant with expertise in PowerShell scripting and command line.\n2. Assume user is using the operating system `osx` unless otherwise specified.\n3. Use the `code block` syntax in markdown to encapsulate any part in responses that is code, YAML, JSON or XML, but not table.\n4. When encapsulating command line code, use '```powershell' if it's PowerShell command; use '```sh' if it's non-PowerShell CLI command.\n5. When generating CLI commands, never ever break a command into multiple lines. Instead, always list all parameters and arguments of the command on the same line.\n6. Please keep the response concise but to the point. Do not overexplain."
      }
  ],
  "Active": "ps-az-gpt4"
}

After that id didn’t expect to have to select the azure agent or to see the name i’ve configured

Summary

You have learned how to install and use the AI Shell in Windows Terminal. Now it’s time to use it and figure out how this fits your work.

Regards
Andres Bohren