Package assistant

The Package assistant is an AI feature that helps you find, create, edit, and review Windows custom packages in PDQ Connect using plain language. Instead of researching silent install switches, success codes, and command-line syntax yourself, you can describe what you want (for example, "Build a package to install Google Chrome and reboot") and the assistant proposes a fully configured package for you to review and approve.

You stay in control the entire time. The assistant drafts and recommends, but nothing is saved to a package or deployed to your devices until you review it and approve it.

 

Early Access: The Package assistant is in Early Access and is free to use while in Early Access. Pricing and availability may change when it becomes generally available.

Note: The Package assistant works with Windows packages only. It does not support macOS packages.

Enable the Package assistant

The Package assistant is part of PDQ Connect's AI tools, which are in Early Access. AI tools are managed org-wide from one place.

  1. Go to Settings > AI tools.
  2. Turn on the Package assistant toggle.

A few things to know about enabling AI tools:

  • AI tools are enabled or disabled org-wide, not per user.
  • You can turn any AI feature on or off at any time, giving you flexibility around your organization's security policies.
  • Enabling and disabling AI features is captured in your Audit logs.

For details on how PDQ handles data for AI features, see PDQ Connect AI Data and Safety FAQ.

Open the Package assistant

The Package assistant lives on the package editing screen for custom packages.

  1. Open an existing custom package, or create a new one from the Packages tab (see Custom Packages).
  2. On the package's Windows view, click the Package assistant button.
  3. The assistant opens in a panel alongside your package. Describe what you want, or pick one of the suggested prompts to get started.

The Package assistant button appears when you are working on an editable Windows package. It is not available on the macOS view of a package or when a package cannot be edited.

What the Package assistant can do

You can use the assistant in three main ways.

Create a package

Describe the software or outcome you want and the assistant will build a package for you. For example:

  • "Create a package to install Google Chrome"
  • "Create a package that cleans up temporary files and frees disk space"
  • "Install Slack, copy our config file, then reboot"
  • "Write a PowerShell script to disable Windows telemetry"

Before it drafts anything, the assistant checks the PDQ Package Library for a matching maintained package. If a good Library match exists, it will tell you and let you choose whether to reference that package (as a Nested step) or build a custom one. For software it needs to install directly, it researches the correct silent install method rather than guessing, then presents a plan and, once you approve it, a proposed package for you to review.

You can also ask for a script-only package, which is especially handy when you need a PowerShell or CMD script but don't want to write it from scratch. Describe what the script should do and the assistant writes it and builds it into a package for you, following safe scripting practices such as error handling and logging that shows up in your Connect deployment output.

Edit a package

With a package open, you can ask the assistant to change it. For example:

  • "Add a reboot at the end"
  • "Change the silent switch on step 1 to /quiet"
  • "Remove the file copy step"
  • "Move the reboot to before the script"

The assistant only edits the package you currently have open, and it presents the proposed changes for your approval before anything is applied. If a change would break the package (for example, removing an install step that a later script depends on), it flags the problem first.

Ask questions and review a package

You can ask the assistant to explain a package or a specific setting without changing anything. For example:

  • "What does this package do?"
  • "Why is the success code 3010 on step 2?" (3010 means success, reboot required.)
  • "How do I make this run as the logged-on user?"

These requests are read-only. The assistant answers your question and does not modify the package.

How a package gets built

When you ask the assistant to create or change a package, it works through a short, guided flow so you always know what is about to happen:

  1. Understand the request. If your request is ambiguous (for example, "WARP" could mean Cloudflare WARP or the Warp terminal), the assistant asks a clarifying question instead of guessing.
  2. Check the Package Library. For install requests, it searches for a matching PDQ-maintained package and lets you choose to use it or build custom.
  3. Research the install. For custom installs, it looks up the correct silent install command, parameters, and success codes rather than relying on memory.
  4. Confirm the plan. It presents a plain-language plan of the steps it intends to build and waits for your go-ahead.
  5. Propose the package. It returns a proposed package for you to review. Read through the steps, and if you want a change, tell the assistant and it will update the proposal. When it looks right, apply it. Nothing is saved until you do.

Supported package steps

The assistant builds packages using the same steps you would add manually. For full detail on each step type, see Custom Packages.

Step What the assistant uses it for
Install Silently install or uninstall an application using the correct command line, parameters, and success codes.
Script Run a PowerShell or CMD script for configuration changes and custom actions.
File Copy Copy files or folders to a target location on the device.
Reboot Restart the device, with a delay and an optional message for the logged-on user.
Nested Reference an existing PDQ Library or custom package as a step, so you reuse work instead of rebuilding it.

What makes a good generated package

When the assistant builds a package, it aims for the same qualities you would want from a package you built by hand:

  • Silent: installs run without pop-ups, wizards, or prompts.
  • Verifiable: install and script steps use the correct success codes so a real success is not reported as an error.
  • Self-cleaning: temporary files copied during a deployment are removed afterward.
  • Runs in the right context: steps use the appropriate Run As mode (Local System or Logged-on User) for what they are doing.
  • Non-disruptive: reboot steps default to a delay and a message so users are not caught off guard.
  • Safe to re-run: packages are built to run more than once without causing problems.

Tips for better results

  • Be specific. Name the exact application and any options you want, such as a version, 64-bit, or disabling auto-updates.
  • Have your details ready. Some software needs values only you have, such as a license key, customer ID, or tenant URL. The assistant will ask for these when they are required.
  • Upload your own installers when needed. For proprietary software with no public download, the assistant will provide guidance and ask you to upload the installer file. It does not attach installer files for you.
  • Answer clarifying questions. When the assistant offers you choices, picking one keeps the build accurate rather than having it guess.

Current limitations

  • Windows only. macOS packages are not supported.
  • One package per session. It can build a package with multiple steps, but not multiple packages at once.
  • You provide the installer files. The assistant suggests download links and asks you to upload files; it does not source or attach installation media for you.
  • It does not deploy for you. The assistant produces a package for you to review and approve. Deploying it to devices is a separate step you control.

Data, safety, and your responsibility

AI output from the Package assistant is advisory. You review and approve every package, step, and script it proposes before it is saved or deployed, and you remain responsible for what you deploy to your devices.

Your prompts and the content the assistant generates are used only to deliver the feature you requested. By contract, PDQ's AI subprocessors do not use your data to train their models, and requests to AI providers are sent over TLS. You can disable AI features at any time from Settings > AI tools, and those changes are recorded in your Audit logs.

For the full details, see PDQ Connect AI Data and Safety FAQ.

Related articles

Was this article helpful?