Engine Mentor

Getting started

1. Enable the plugin under Edit → Plugins → Engine Mentor and restart the editor. The engine plugin Toolset Registry (experimental) is enabled automatically.
2. Open the Mentor: Window → Engine Mentor or Ctrl+Shift+M.
3. The first time you open it, the setup walks you through language, way to AI, voice and privacy. You can skip it and restart it later via → Run setup again at the top of the Mentor.

Keyboard shortcuts

AI providers

The Mentor needs an AI for the chat. There are two ways – and the verified recipes work without any AI at all.

Setting up Claude Code

1. Settings → AI provider (or in the setup): Install Claude Code. A console window opens with Anthropic's official installer (irm https://claude.ai/install.ps1 | iex). It verifies the checksum and installs Claude Code for your Windows user.
2. “On PATH” should show yes. Otherwise click “Add to PATH” – then claude also works in any console.
3. Sign in opens Claude Code in a console. Type /login there and follow the steps in the browser, then /exit.

The copy of Claude Code bundled with Claude Desktop is found as well. It changes with every Claude Desktop update and is not on the PATH, so the official installer is the better choice. You can also pick the path yourself (Choose path …).

For Claude you can additionally choose Claude account (subscription) or Own API key (Anthropic, billed per use). If ANTHROPIC_API_KEY is set in your environment, Claude Code bills through it – the Mentor points this out.

ChatGPT via the Codex CLI

Install Codex CLI starts OpenAI's official installer (irm https://chatgpt.com/codex/install.ps1 | iex). Then click Sign in (opens codex login) and confirm with your ChatGPT account. In the Mentor the Codex CLI runs read-only.

API key

Pick the provider, enter your key and save. “Check” only fetches the model list and costs nothing. Keys are stored encrypted with Windows DPAPI; environment variables (e.g. OPENAI_API_KEY, XAI_API_KEY, GEMINI_API_KEY, MISTRAL_API_KEY, OPENROUTER_API_KEY) work too. The screenshot needs a model with image understanding.

Difference to Claude: API-key providers get no tools and read no project files – only the screenshot, the list of controls and your text. You can switch providers in the middle of a conversation.

The four modes

Only a short version is read aloud: one or two sentences to get started – the full answer is in the chat. Steps appear one at a time in the chat and are announced one at a time as you reach them; you only hear step 2 once step 1 is done. Finished steps get a ✓. Keys are shown as keycaps, e.g. Ctrl+S. When you are meant to try keys yourself (such as W A S D in the viewport), each key you press lights up green in the step card; once all have been pressed, it continues on its own. Typing in the chat does not count.

If the guide cannot find a target or you click somewhere else, the Mentor adjusts automatically (setting Adjust automatically when steps do not match).

Recipes and lessons

→ Recipes at the top of the Mentor opens 250 common tasks in 9 areas. Show or Demonstrate starts a recipe. If there is a verified path for your Unreal version and editor language, the Mentor replays it directly – without AI. Otherwise the Mentor checks the way with the AI and guides you.

• When a guide runs through completely, the recipe counts as verified for your version and is saved locally.
• If a replayed path fails, it is marked as “changed” and verified again next time.
• Status: verified · verified in another version · changed · open · explanation.

Recipe recognition

If your question matches a recipe, the Mentor suggests it. Recognition runs locally on your PC (a small language model; nothing leaves your machine).

Lessons

The start card offers lessons to try. Put your own lessons as JSON under %LOCALAPPDATA%\EngineMentor\Lessons\<language>\. Every completed guide goes into the learning log, and the Mentor builds on it.

Voice and voice input

Read-aloud (Piper) and voice input (Whisper) run locally on your PC. For this the Mentor downloads packages when you click – in the setup or under Settings → Sidecars → Set up automatically:

The Mentor verifies every file against the SHA-256 checksum built into the plugin and only then unpacks it – to %LOCALAPPDATA%\EngineMentor\Sidecars\, once per Windows user for all projects (you can change the folder, see Privacy and storage). Without the download the Windows voice reads aloud and voice input stays off.

Dictation

Microphone button or Ctrl+Shift+Space. The microphone stays on; at every pause the segment is recognised and appended. A second click ends it; after 20 s without speech the microphone turns off by itself.

Technical terms

Unreal terms come out right even with a strong accent. Maintain your own terms under Settings → Speech → Your terms; you can also record and test them there. The Train terms window goes through the bundled terms by area.

Playing and finding errors

While you play in the editor (Play), the Mentor listens to the log. It announces new errors and Blueprint runtime errors such as Accessed None trying to read property … right away – locally with the voice, without AI and at no cost. At the same time it asks the AI for the cause and reads the explanation to you while you are still playing.

Overlay in the game window

The Mentor icon sits in the bottom right of the game window. When there is an error, it shows the count and the shortcut. Ctrl+Shift+M opens the overlay while playing: it shows the error and the explanation, and the Mentor reads them aloud. Press it again to close it. The overlay never captures mouse or game input; Shift+F1 releases the mouse as usual.

Find them under Settings → General → Play mode · Live debugger. With a subscription an explanation costs nothing extra; with an API key your provider bills every request.

Play log

All warnings and errors of the last play session are in Saved\EngineMentor\PlayLog.md – identical messages merged and counted, Blueprint errors with Blueprint, function and node. After playing, just ask (“What was that?”); the Mentor receives the new entries automatically.

Project overview

The Mentor reads the names of your levels, Blueprints (with parent class), C++ classes and assets once and keeps them up to date (Saved\EngineMentor\ProjectIndex.md). That way it knows right away what your player Blueprint or GameMode is called and does not have to search first – faster answers, fewer tokens. You can turn it off under Settings → General → Project overview.

Settings

Gear icon at the top of the Mentor. Rare options are under All options in Editor Preferences … (Editor Preferences → Plugins → Engine Mentor).

At the top of the Mentor, Help opens this page online or locally. Under you find recipes, Run setup again and detaching or docking the window. Settings → General → Setup resets the guided setup; your settings are kept.

Privacy and storage

• Web search: if allowed (default), Claude or Codex may look things up online when unsure about your Unreal version.
• Local: read-aloud, voice input and recipe recognition run on your PC. Packages are downloaded only when you click.
• Project and playing: the project overview (names of levels, Blueprints, classes and assets) and error messages from play mode go to your AI provider as a short summary or when needed. You can turn both off in the settings.
• In the project (Saved\EngineMentor\): chat histories, system text, last screenshot.
• For your Windows user (%LOCALAPPDATA%\EngineMentor\): encrypted API keys (DPAPI), sidecars, learning log, your lessons and terms, your verification runs.

You can change this data folder under Settings → Sidecars → Data folder, for example to another drive. The Mentor copies everything to the new folder; the old one stays until you delete it yourself.

More in the privacy notice.

Troubleshooting

“Claude Code not found”

Install Claude Code or Choose path …, then Search again. Without Claude Code, Claude works neither by subscription nor by Anthropic key – choose a provider with an API key instead.

Program no longer starts / Node.js removed

Reinstall runs the official installer again; sign-in and settings are kept. An older Codex CLI from npm needs Node.js – the Mentor marks it and offers Install standalone (no Node.js). Then Search again.

Not signed in / subscription not recognised

Sign in and type /login in the console (Claude) or confirm in the browser (Codex). The status shows the subscription as soon as the CLI confirms the sign-in.

Invalid key

“Check” shows the provider's message. Often the key ID was copied instead of the key; the Mentor points out an unusual format.

Voice or voice input does not download

If the server cannot be reached, download the packages again later or choose an unpacked package folder yourself under Sidecars. Interrupted downloads resume. A file with a wrong checksum is discarded and not unpacked.

“Data folder not readable” / settings seem reset

Sometimes Windows denies an editor process access to the data folder, for example after restarting the PC. Sidecars, terms or training progress then look as if they were gone. The data is not lost, and the Mentor overwrites nothing in this state. Closing and restarting the editor usually helps. If it happens often, choose another data folder.

The Mentor cannot find a control

Search again on the step card or Ask the Mentor. With adjust automatically on, the Mentor adapts the remaining steps itself. Menus and labels depend on the Unreal version and editor language.

Log

In the editor's Output Log under the category LogEngineMentor. The file is Saved\Logs\<Project>.log.

Third-party software and licenses

The full list with licenses and links is under Settings → About. In short:

• Icons: Lucide (ISC); provider logos: LobeHub Icons (MIT). Logos are trademarks of their owners and are used for identification only.
• Fonts of this help: Inter, Space Grotesk and JetBrains Mono (SIL OFL 1.1, licenses in the folder Docs\Help\_shared\fonts).
• Recipe recognition: model multilingual-e5-small (MIT), run with the engine plugin NNERuntimeORT (Beta).
• Read-aloud: Piper (GPL-3.0, separate download, source code included in the package); voices de_DE-mls-medium (CC BY 4.0) and en_US-ljspeech-medium (public domain).
• Voice input: faster-whisper and CTranslate2 (MIT), Whisper models (MIT), NVIDIA runtime libraries (NVIDIA license).
• Claude, ChatGPT, Codex, Grok, Gemini, Mistral and OpenRouter are trademarks of their owners. Unreal Engine is a trademark of Epic Games.

Help and contact

This help matches your version 0.5.0. For newer notes, answers and contact: