Let’s Embed AI in PowerShell. Part 1. PS + Gemini-CLI

In this experiment, I use Gemini-CLI as a search model.

What is Gemini CLI?

I have already talked in detail about **Gemini CLI** in Gemini CLI: Introduction and First Steps. But if you missed it, here’s a brief introduction.

In short, **Gemini CLI** — is a command-line interface for interacting with Google‘s AI models. You run it in your terminal, and it turns into a chat that, unlike web versions, has access to your file system.

Key features:

• Understands code: It can analyze your scripts, find errors in them, and suggest fixes.
• Generates code: You can ask it to write a PowerShell script to solve your problem, and it will.
• Works with files: Can read files, create new ones, and make changes to existing ones.
• Runs commands: Can execute shell commands, such as git or npm.

For our purposes, the most important thing is that Gemini CLI can work in **non-interactive mode**. That is, we can pass it a prompt as a command-line argument, and it will simply return an answer, without launching its interactive chat. This is the capability we will use.

**Installation and Setup**

To get started, we need to prepare our environment. This is done once.

Step 1: Install Node.js
Gemini CLI — is an application written in Node.js (a popular JavaScript runtime environment). So first, we need to install Node.js itself.

1. Go to the official website: https://nodejs.org/
2. Download and install the **LTS** version. This is the most stable and recommended option. Just follow the installer instructions.
3. After installation, open a new PowerShell window and check that everything is working: node -v npm -v You should see versions, for example, v20.12.2 and 10.5.0.

Step 2: Install Gemini CLI itself
Now that we have npm (the package manager for Node.js), installing Gemini CLI comes down to one command. Run it in PowerShell:

npm install -g @google/gemini-cli

The -g flag means “global installation,” which will make the gemini command available from anywhere in your system.

Step 3: Authentication
The first time you launch Gemini CLI, it will ask you to sign in to your Google account. This is necessary so that it can use your free quota.

1. Simply type the command in PowerShell: gemini
2. It will ask you about signing in. Select “Sign in with Google.”
3. In your browser will open a standard Google sign-in window. Sign in to your account and grant the necessary permissions.
4. After that, you will see a welcome message from Gemini in the console. Congratulations, you are ready to work! You can type /quit to exit its chat.

The Terrible Invoke-Expression

Before we put it all together, let’s get acquainted with one of the most dangerous cmdlets in PowerShell — Invoke-Expression, or its short alias iex.

Invoke-Expression takes a text string and executes it as if it were a command typed in the console.

Example:

$commandString = "Get-Process -Name 'chrome'"
Invoke-Expression -InputObject $commandString

This command will do the same as a simple call to Get-Process -Name 'chrome'.

Why is it dangerous? Because executing a string that you don’t control (for example, obtained from the internet or from AI), — is a huge security hole. If the AI, by mistake or with malicious intent, returns the command Remove-Item -Path C:\ -Recurse -Force, iex will execute it without hesitation.

For our task — creating a managed and controlled bridge between a natural language request and its execution — it is perfectly suited. We will use it with caution, fully aware of the risks.

Putting It All Together: The Invoke-Gemini Cmdlet

Let’s write a simple PowerShell function that will allow us to send prompts with a single command.

Copy this code and paste it into your PowerShell window to make it available in the current session.

function Invoke-Gemini {
    <#
    .SYNOPSIS
        Sends a text prompt to Gemini CLI and returns its response.
    #>
    [CmdletBinding()]
    param(
        [Parameter(Mandatory = $true, Position = 0, ValueFromPipeline = $true)]
        [string]$Prompt
    )

    process {
        try {
            # Check if the gemini command is available
            $geminiCommand = Get-Command gemini -ErrorAction Stop
        }
        catch {
            Write-Error "Command 'gemini' not found. Make sure Gemini CLI is installed."
            return
        }

        Write-Verbose "Sending prompt to Gemini CLI..."

        # Run gemini in non-interactive mode with our prompt
        $output = & $geminiCommand.Source -p $Prompt 2>&1

        if (-not $?) {
            Write-Warning "Gemini command finished with an error."
            $output | ForEach-Object { Write-Warning $_.ToString() }
            return
        }

        # Return clean output
        return $output
    }
}

Trying the Magic!

Let’s ask it a general question directly from our PowerShell console.

Invoke-Gemini -Prompt "Tell me about the five latest trends in machine learning"

Congratulations! You have just successfully embedded AI in PowerShell.

In the next article, I will tell you how to use Gemini CLI to run scripts and automate tasks.