Vim grammar checker for markdown files, powered by AI. Supports any OpenAI API compatible model, local, remote, Claude and AWS Bedrock.

Use your plugin manager to install vim-gramaculate

Plug 'ahalbert/vim-gramaculate'

The defaults assume you want to use a locally run LLM to check grammar. Install Ollama and then install the model qwen3:4b. If you want to use a different model, set g:gramaculate_model.

It's a lot faster to call a model such as Claude Haiku on an underpowered machine. Set the following variables to call another model:

let g:gramaculate_url       = 'https://api.anthropic.com/v1/messages'
let g:gramaculate_model     = 'claude-haiku-4-5-20251001'
let g:gramaculate_api_key   = 'YOUR-API-KEY'

Set g:gramaculate_url to your Bedrock endpoint and g:gramaculate_model to the Bedrock model ID. The region is extracted from the URL automatically. AWS credentials are read from the environment or ~/.aws/credentials via the aws CLI, which must be installed and configured.

let g:gramaculate_url         = 'https://bedrock-runtime.us-east-1.amazonaws.com'
let g:gramaculate_model       = 'anthropic.claude-haiku-4-5-20251001-v1:0'
let g:gramaculate_aws_profile = 'my-profile'  " optional, defaults to AWS default profile

Use :Gramaculate to check the grammar of an entire markdown file. vim-gramaculate supports ranges, so you can select text in visual mode and only check it. Once the model returns the fixes, you can use <Enter> in the window to apply a change, or a mapping you set in g:gramaculate_map_apply (default: <leader>cg) to apply the change in the buffer. Use 'q' in the vim-gramaculate window to close it and clear the highlights.

I made gramaculate because I wasn't getting useful fixes from more traditional grammar checkers, but you may find them useful because they run much faster and integrate with ALE. Unfortunately, checking on save is cost and performance prohibitive (for now).

• Harper
• markdown-lint
• Vale
• proselint
• write-good