Extension:WandaScribe

**MediaWiki extensions manual**
WandaScribe; Release status: stable
Implementation	Artificial intelligence
Description	WandaScribe is a MediaWiki extension that provides literary assistance by utilizing the LLM capability of the Wanda. It helps users write better content in the edit window with features like spell check, grammar check, text improvement, and more.
Author(s)	Naresh Kumar Babu (TechieNKtalk)
Maintainer(s)	Techwizzietalk
Latest version	1.0.0
MediaWiki	1.42+
	Hooks used BeforePageDisplay;
Licence	MIT License
Download	Download extension ; Git [?]: Browse repository (GitHub); Gerrit code review; Git commit log; Download source tarball;
	Translate the WandaScribe extension if it is available at translatewiki.net
Issues	Open tasks · Report a bug

WandaScribe is a MediaWiki extension that provides literary assistance by utilizing the LLM capability of the Wanda. It helps users write better content in the edit window with features like spell check, grammar check, text improvement, and more.

Installation

Download and move the extracted WandaScribe folder to your extensions/ directory.
Developers and code contributors should install the extension from Git instead, using:
```
cd extensions/
git clone https://gerrit.wikimedia.org/r/mediawiki/extensions/WandaScribe
```
Add the following code at the bottom of your LocalSettings.php file:
```
wfLoadExtension( 'WandaScribe' );
```
Ensure that Wanda extension is installed.
Done – Navigate to Special:Version on your wiki to verify that the extension is successfully installed.

Features

Real-time Assistance

Spell Checking: Real-time detection of misspelled words with AI-powered suggestions
Grammar Checking: Identifies grammatical errors and suggests corrections
Text Improvement: AI-powered suggestions to enhance clarity and engagement
Tone Adjustment: Convert text between formal and casual tones
Text Simplification: Make complex text easier to understand
Text Expansion: Add more details and context to brief text
Summarization: Create concise summaries of longer text

User Interface

Assistance Panel: Integrated panel in the editor with easy-to-use buttons
Non-intrusive Popup: Suggestions appear in a responsive popup near text selection
Responsive Design: Works on desktop, tablet, and mobile devices
Modern UI: Built with Wikimedia's Codex design system

Configuration

WandaScribe requires the Wanda extension to be installed and configured with an LLM provider. No additional configuration is required for WandaScribe itself.

Wanda Configuration

Make sure Wanda is properly configured in your LocalSettings.php:

// Example configuration for local Ollama
$wgWandaLLMProvider = 'ollama';
$wgWandaLLMModel = 'gemma:2b';
$wgWandaLLMApiEndpoint = 'http://localhost:11434/api/';

// Example configuration for OpenAI
$wgWandaLLMProvider = 'openai';
$wgWandaLLMModel = 'gpt-3.5-turbo';
$wgWandaLLMApiKey = 'your-api-key-here';

For detailed Wanda configuration options, see Wanda's configuration documentation.

Usage

Basic Workflow

Navigate to any wiki page and click "Edit" to open the wikitext editor
The WandaScribe assistance panel appears above or near the editor
Select text in the editor that you want to improve
Click one of the assistance buttons (e.g., "Check Spelling", "Improve Text")
Review the AI-generated suggestion in the popup
Click "Apply" to accept the suggestion or "Dismiss" to reject it

Available Actions

Button	Description	Use Case
Check Spelling	Identifies misspelled words and provides corrections	Finding typos and spelling errors
Check Grammar	Detects grammatical errors and suggests fixes	Improving sentence structure and grammar
Improve Text	Enhances clarity, engagement, and overall quality	General text improvement
Make Formal	Converts text to a formal tone	Professional documentation
Make Casual	Converts text to a casual tone	Friendly, conversational content
Simplify	Makes complex text easier to understand	Simplifying technical content
Expand	Adds more details and context	Elaborating on brief descriptions
Summarize	Creates concise summaries	Condensing long text

Privacy and Security

Data Processing

All text processing is performed through the Wanda extension's LLM integration
Selected text is sent to the configured LLM provider (Ollama, OpenAI, etc.)
No text is stored by WandaScribe itself

Privacy Considerations

For public wikis: Users should be aware that selected text is sent to external LLM providers (unless using self-hosted Ollama)
For private wikis: Consider using self-hosted Ollama to keep all data on-premises
For sensitive content: Use local LLM providers to ensure data privacy

Permissions

WandaScribe inherits all permissions from the edit page
Users must have edit permissions to see the assistance panel
No additional permissions are required

Troubleshooting

Panel doesn't appear

Verify you're on an edit page (action=edit)
Check that Wanda extension is installed and loaded at Special:Version
Check browser console for JavaScript errors
Ensure JavaScript is enabled in the browser

"Wanda not available" error

Verify Wanda extension is installed: wfLoadExtension( 'Wanda' );
Check Wanda configuration in LocalSettings.php
Verify LLM provider is properly configured and accessible
Test Wanda's chat feature at Special:Wanda

No suggestions or slow responses

Check Wanda's LLM configuration and endpoint
Verify the LLM service is running and accessible
Review MediaWiki debug logs for errors
Consider using a faster LLM model or local Ollama instance

Popup positioning issues

The popup uses responsive positioning and should adapt to screen size
Try refreshing the page if positioning seems incorrect
Check browser console for JavaScript errors

Performance Considerations

Response time depends on the configured LLM provider
Local Ollama instances typically provide faster responses than cloud providers
Smaller LLM models (e.g., gemma:2b) are faster but may provide lower quality suggestions
Larger models (e.g., gpt-4) provide better quality but are slower and may incur costs

Limitations

Requires Wanda extension to be installed and configured
Quality of suggestions depends on the LLM model used
Response times vary based on LLM provider
No offline mode (requires active LLM connection)