- \n
- Improved support for Azure PowerShell<\/li>\n
- Broader support for third party OpenAI-compatible models<\/li>\n
- Refactored code sample for creating an Ollama agent<\/li>\n
- Improved native command error handling<\/li>\n
- An easier way to deploy Azure OpenAI instances<\/li>\n
- Additional bug fixes<\/li>\n<\/ul>\n
<\/i>AI Shell Preview 3 is released!<\/strong><\/p>Preview 3 has been released just two weeks after Preview 2 due to a backend issue affecting the Azure Agent. This update addresses the problem, ensuring more reliable interactions with Azure services.<\/div><\/p>\n
Improved support for Azure PowerShell<\/h2>\n
The improvements to the
azure<\/code> agent are designed to reduce friction when managing Azure resources\nwith the Azure CLI and PowerShell. The agent now offers deeper integration with Azure PowerShell,\nmaking it easier to use AI-created Azure PowerShell scripts.<\/p>\n- \n
- The
\/replace<\/code> command now supports Azure PowerShell. The agent walks you through replacement of\nparameter values in generated Azure PowerShell responses.<\/li>\n- The agent now support Azure PowerShell authentication using the
Connect-AzAccount<\/code> command.\nPreviously, the agent only supported Azure CLI authentication.<\/li>\n<\/ul>\nBroader support for third party OpenAI-compatibles models<\/h2>\n
With this release, the
openai-gpt<\/code> agent extends support to third party models that follow the\nOpenAI API specifications, allowing for a more flexible AI experience. Many of these models are open\nsource tools for running SLMs and LLMs locally. Theopenai-gpt<\/code> agent nows supports the following\nadditional models:<\/p>\n- \n
- Ollama<\/strong><\/a><\/li>\n
- LM Studio<\/strong><\/a><\/li>\n
- Deepseek<\/strong><\/a><\/li>\n
- LocalAI<\/strong><\/a><\/li>\n
- Google Gemini<\/strong><\/a><\/li>\n
- Grok<\/strong><\/a><\/li>\n<\/ul>\n
To use these models, you only need to configure the
endpoint<\/code>,key<\/code> andmodel name<\/code> in the agent\nconfig file. The followingopenai.agents.config<\/code> file contains example configurations for two of\nthe new AI providers.<\/p>\n{\r\n \"GPTs\": [\r\n {\r\n \"Name\": \"gpt-deepseek\",\r\n \"Description\": \"A GPT instance using DeepSeek v3.\",\r\n \"Endpoint\": \"https:\/\/api.deepseek.com\",\r\n \"ModelName\": \"deepseek-chat\",\r\n \"Key\": \"<your-deepseek-api-key>\",\r\n \"SystemPrompt\": \"You are a helpful assistant.\"\r\n },\r\n\r\n {\r\n \"Name\": \"gpt-gemini\",\r\n \"Description\": \"A GPT instance using Google Gemini.\",\r\n \"Endpoint\": \"https:\/\/generativelanguage.googleapis.com\/v1beta\/openai\/\",\r\n \"ModelName\": \"gemini-1.5-flash\",\r\n \"Key\": \"<your-gemini-api-key>\",\r\n \"SystemPrompt\": \"You are a helpful assistant.\"\r\n }\r\n ]\r\n\r\n \"Active\": \"gpt-deepseek\"\r\n}<\/code><\/pre>\nImproved native command error handling<\/h2>\n
Resolve-Error<\/code> (aliasfixit<\/code>) now uses the screen scraper API on Windows to capture the error\noutput of native commands. This allows AI Shell to provide better resolution of error messages that\ncome from native commands rather than from PowerShell.<\/p>\n![\"Resolve](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/02\/native-command-output.png\")
<\/p>\nOllama agent improvements<\/h2>\n
We refactored the example code for the agent
ollama<\/code>. This example provides the basic structure\nthat you can use to build out your own agent. The refactored code now uses the OllamaSharp<\/strong>\nlibrary and add the ability to configure the agent using settings files.<\/p>\nThe
ollama<\/code> agent isn’t shipped with AI Shell but you can build it yourself. See the instructions\nin the locally build the repo<\/a> section of the README. Thank you@kborowinski<\/code> on GitHub for\nthis contribution!<\/p>\nTools for easy deployment of Azure OpenAI instances<\/h2>\n
We added a Bicep template that deploys an Azure OpenAI instance in seconds. Using the template is\nmuch faster than manually creating an instance in the Azure portal. This template includes\ndocumentation with step by step instructions.<\/p>\n
- \n
- Bicep template<\/a><\/li>\n
- Documentation<\/a><\/li>\n<\/ul>\n
Additional bug fixes<\/h2>\n
There are many bug fixes in this release. For a complete list of changes, see the changelog<\/a>.<\/p>\n
How to get AI Shell Preview 2<\/h2>\n
Use the following command in PowerShell 7 to install or upgrade AI Shell:<\/p>\n
Invoke-Expression \"& { $(Invoke-RestMethod 'https:\/\/aka.ms\/install-aishell.ps1') }\"<\/code><\/pre>\nFor more information, see Install AI Shell<\/a>.<\/p>\n
Automatically start AI Shell<\/h2>\n
One piece of feedback we often hear is: How can I have AI Shell automatically start up when I open\nWindows Terminal?<\/em><\/p>\n
It’s easy! Just add
Start-AIShell<\/code> to your PowerShell profile script. Here’s how:<\/p>\n- \n
- Open your profile in the text editor of your choice.<\/li>\n
- Add the following code to the end of the file:\n
if ($PSVersionTable.PSVersion -ge ([version]'7.4.6') -and\r\n (Get-Process -Id $pid).Parent.Name -eq 'WindowsTerminal') {\r\n Start-AIShell\r\n}<\/code><\/pre>\n<\/li>\n<\/ol>\nNow, AI Shell will automatically start when you open Windows Terminal.<\/p>\n
<\/i>NOTE<\/strong><\/p>AI Shell only works with Windows Terminal and requires PowerShell 7.4.6 or higher.<\/div><\/p>\n
If you need help creating a profile scripts, see about_Profiles<\/a>.<\/p>\n
Try it out and share your feedback<\/h2>\n
- Documentation<\/a><\/li>\n<\/ul>\n
- LM Studio<\/strong><\/a><\/li>\n
- The agent now support Azure PowerShell authentication using the
- The