WinToolify

EN TR

Windows Management Harness

WinToolify gathers the settings, services, apps and repair commands Windows scatters across a dozen places onto one keyboard-driven terminal. Every row states its risk, every change is recorded so you can undo it, and the same catalogue is open to an AI assistant.

Works on Windows 10 and 11 · Works on PowerShell 5.1 and 7 · MIT

irm https://github.com/burakarslan0110/WinToolify/releases/latest/download/WinToolify.ps1 | iex

Runs with administrator rights.

A recording of the console. On the Windows Services screen, DiagTrack and the geolocation service are each marked in turn and then applied, and both rows change from running to stopped and disabled. On the Privacy and Telemetry screen, three settings are marked one at a time and applied together. The Undo Last Change screen then holds both of those applies as records. It ends in the assistant: asked what caused the last blue screen, it reads the crash topic, searches the web, reads Microsoft's page on the stop code, and offers the matching report as suggestion one, which runs when its number is typed.

  • 51repair and maintenance commands
  • 50read-only reports
  • 43Windows services
  • 242privacy and system settings
  • 122preinstalled apps
  • 252programs in the store
  • 7assistant tools, none of which can change anything

Five moments from the interface

Every screen is the same screen

The main menu box: twelve numbered rows, a description band, and the navigation guide.

Seven of the 43 services, each with a risk tag and its current status and start type. BDESVC, the BitLocker service, is tagged ADVANCED and drawn in red.

A privacy screen with three rows marked. Their state column has switched from "Not applied" to a yellow "Will apply", and the footer now offers Enter to apply.

The undo screen: four recorded changes, newest first, each with its timestamp, what it did and how many items it covered.

The assistant answering "my pc got slow, have a look": three read-only tool calls with their timings, a short diagnosis, then two numbered suggestions with their risk tags and where they live in the menus.

AI Assistant Harness

Reads everything, touches nothing

WinToolify's catalogue is a list of things that can be done, and there are two ways to reach it. You can walk the menus yourself, or you can ask the AI assistant. Both end up at the same entries, the same confirmations and the same undo records.

The seven tools the AI can call

  • search_wintoolifySearches the catalogue, meaning settings, tools, screens and store apps.
  • suggest_wintoolifyOffers what it found as a numbered list you pick from.
  • read_systemReads one topic of this machine: the overview, errors, crashes, storage, network, security, startup, services, processes, or what WinToolify changed.
  • web_searchnetworkMicrosoft Learn first, then DuckDuckGo.
  • fetch_pagenetworkReads one of those results as text.
  • read_rawOne registry key, or the first lines of one file, with credential stores refused.
  • save_noteRemembers one short fact for next time.

Seven tools, all of them read only

There is no eighth tool that applies anything. The assistant prints a numbered card and stops, and you type the number.

Your endpoint, your machine

Anything OpenAI-compatible works, OpenAI’s own API included, so ChatGPT models sit alongside LM Studio and Ollama.

What it never gets to see

Computer name, user names, profile paths, BIOS serial, MAC and IP addresses become fixed tokens before any result reaches the model. There is no off switch.

It looks things up

When the problem is newer than the model, it searches Microsoft Learn and then the web, and cites the URL it used.

The same assistant transcript shown in the walkthrough, at a readable size.

Safety model

What does a change go through?

  1. Step 1

    Read

    WinToolify reads the current state of every marked item.

  2. Step 2

    Record

    That state goes into an undo record on disk, before anything is applied.

  3. Step 3

    Apply

    Each item is changed, one at a time.

  4. Step 4

    Verify

    The state is read again and reported per row, because a cmdlet can succeed while policy or a pending reboot leaves the item unchanged.

  • Every row is tagged SAFE, CAUTION or ADVANCED, and anything above SAFE spells out what stops working.
  • An ADVANCED row will not take a keypress. You have to type the word out: YES in English, EVET in Turkish.
  • A row that does not apply to your build of Windows is left out of the list rather than shown greyed out.
  • Removing a setting puts the Windows default back. Where no default is known, the row cannot be removed at all.
  • Start it without administrator rights and it relaunches itself elevated.
  • Nothing reaches the disk until you apply something.

The typed confirmation panel for an ADVANCED row: the consequence in red, the change about to be made, and a footer asking the reader to type YES.

An ADVANCED row does not take a keypress for an answer. The consequence is stated first, then the word has to be typed.

Get started

How do you run it?

What you need

  • Operating systemWindows 10 or 11
  • PowerShellWindows PowerShell 5.1, or PowerShell 7
  • RightsAdministrator, or it elevates itself
  • Optionalwinget, for the store
  • OptionalAn OpenAI-compatible endpoint, for the assistant

Run the latest release

Pulls the latest release into memory and starts it. Add -Assistant to open straight into the assistant.

irm https://github.com/burakarslan0110/WinToolify/releases/latest/download/WinToolify.ps1 | iex

Keep the file instead

Download WinToolify.ps1 from the latest release and run it yourself.

powershell -NoProfile -ExecutionPolicy Bypass -File .\WinToolify.ps1

Check what you downloaded

Every release ships a SHA-256 hash and a signed build attestation.

(Get-FileHash .\WinToolify.ps1 -Algorithm SHA256).Hash.ToLower()
gh attestation verify .\WinToolify.ps1 --repo burakarslan0110/WinToolify

Build from source

No modules, no network. The build concatenates the numbered layers under src/ into dist/WinToolify.ps1.

.\build.cmd

Run the checks

Builds, parses, checks 5.1 and 7.0 compatibility, then runs close to 3,000 Pester tests. CI runs exactly this and nothing else.

.\tools\Invoke-WtChecks.ps1