After Grammarly disabled its API, no equivalent grammar-checking tool exists for VSCode. While LTeX catches spelling mistakes and some grammatical errors, it lacks the deeper linguistic understanding that Grammarly provides.
This extension bridges the gap by leveraging large language models (LLMs). It chunks text into paragraphs, asks an LLM to proofread each paragraph, and highlights potential errors. Users can then click on highlighted errors to view and apply suggested corrections.
Features
- LLM-powered grammar checking in American English
- Inline corrections via quick fixes
- Choice of models: Use a local
llama3.2:3b model via Ollama or gpt-40-mini through the VSCode LM API
- Rewrite suggestions to improve clarity
- Synonym recommendations for better word choices
- Configurable system prompts to customize language variant, writing style, and behavior
- Configurable Ollama models to use any local model that fits your needs and hardware
Commands
When the first command is executed, a dialog appears allowing users to select either a local Ollama model or the GitHub Copilot model.
Available Commands
- "LLM Writing Tool: Start Text Check for Current Document"
Continuously checks the text in the current document. Prompts the user to select an LLM model.
- "LLM Writing Tool: Stop Text Check for Current Document"
Stops real-time grammar checking.
- "LLM writing tool: Rewrite current selection"
Rewrites the selected text for clarity.
- "LLM writing tool: Get synonyms for selection"
Suggests synonyms for the selected expression.
- "LLM writing tool: Select model"
Selects the LLM model to use for grammar checking. Stops real-time grammar checking if it is running.
- "LLM writing tool: Reset prompts to defaults"
Resets all customized system prompts back to their default values.
- "LLM writing tool: Reset all settings to defaults"
Resets all extension settings (prompts and Ollama model) back to their default values.
Configuration
The extension now supports configurable system prompts and Ollama model selection, allowing you to customize how the LLM interacts with your text. This enables you to:
- Change the language variant (e.g., British English instead of American English)
- Adjust the writing style (e.g., formal vs. casual tone)
- Use different Ollama models (e.g., faster or higher-quality models)
- Modify the number of synonyms returned
- Customize the behavior for specific use cases
Accessing Settings
- Open VS Code Settings (
Cmd+, on macOS, Ctrl+, on Windows/Linux)
- Search for "LLM Writing Tool"
- Configure the three available prompt settings:
Available Settings
Prompt Configuration:
lmWritingTool.prompts.proofreading
Controls how the extension checks for grammar and spelling errors.
Default: Checks for American English grammar and spelling mistakes.
lmWritingTool.prompts.rewrite
Controls how the extension rewrites text for clarity.
Default: Rewrites text for clarity in American English.
lmWritingTool.prompts.synonyms
Controls how the extension finds synonyms for selected expressions.
Default: Provides up to 5 synonyms.
Ollama Configuration:
lmWritingTool.ollama.model
Specifies which Ollama model to use for local text processing.
Default: llama3.2:3b
Examples: llama3.2:1b, llama3.1:8b, codellama:7b, mistral:7b
Placeholders
When customizing prompts, use these placeholders:
{text} - The text to be processed (for proofreading and rewrite prompts){expression} - The selected expression (for synonyms prompt)
Example Customizations
British English proofreading:
Proofread the following message in British English. If it is grammatically correct, just respond with the word "Correct". If it is grammatically incorrect or has spelling mistakes, respond with "Correction: ", followed by the corrected version. Use British spelling and grammar conventions.\n{text}
Formal writing style:
Rewrite the following text in a formal, academic tone using British English. Maintain the original meaning while improving clarity and formality:\n{text}
More synonyms:
Give up to 10 synonyms for the expression "{expression}". Provide varied alternatives including formal and informal options. Just respond with the synonyms, separated by newlines.
Using a different Ollama model:
Set lmWritingTool.ollama.model to llama3.1:8b for better quality (but slower) processing, or llama3.2:1b for faster (but lower quality) processing.
Resetting Settings
Reset prompts only:
- Open the Command Palette (
Cmd+Shift+P / Ctrl+Shift+P)
- Type "LLM writing tool: Reset prompts to defaults"
- Press Enter
Reset all settings (prompts + Ollama model):
- Open the Command Palette (
Cmd+Shift+P / Ctrl+Shift+P)
- Type "LLM writing tool: Reset all settings to defaults"
- Press Enter
Installation
- Install the extension from the VSCode Marketplace.
- Install Ollama and pull
llama3.2:3b for local grammar checking, or subscribe to GitHub Copilot for online LLM access.
How It Works
- The extension splits the text into sections and sends them to the selected LLM for proofreading.
- It then compares the LLM’s suggestions with the original text to detect changes.
- Detected errors are highlighted, and users can apply quick fixes with a click.
- Responses are cached to minimize repeated API calls.
- Every 5 seconds, the extension checks for text changes and reprocesses modified sections.
Roadmap
- [ ] On-disk caching to improve startup times and reduce redundant API requests.
- [ ] Smarter text chunking to ensure uniform section sizes (e.g., ~2 full lines per section instead of splitting by line).
- [ ] Support for additional languages, starting with British English. Future versions may support any language available in the LLM.
- [ ] Evaluation of alternative models for improved results, with prompt adjustments as needed.
Contributing
Contributions are welcome! Feel free to:
