New open-source LLM plugin auto-generates inclusive copy and accessibility annotations for tokens

AI · 5 min read

An open-source plugin combining a lightweight local LLM with token managers (Style Dictionary, Theo) was released this week, targeting an overlooked part of design systems: the text and semantics that travel with tokens. The plugin suggests inclusive microcopy for labels, placeholder text and error messages and creates accompanying accessibility annotations—like expected ARIA attributes and keyboard behavior notes—that can be attached to tokens.

Developers and designers can run the plugin as part of token publishing pipelines. It reads token metadata and component usage examples, prompts the model with contextual rules (e.g., 'avoid gendered language', 'provide alternatives for color-coded information'), and returns candidate copy and annotation stubs. The output is editable and comes with a confidence score and citation of which guideline was used to make the suggestion.

Early adopters report that the tool saves time on content work and helps surface inclusion issues faster, while contributors stress the need for human review—models still produce culturally biased or undefined phrasing without guardrails. The project includes a policy module that allows teams to bake their cultural, legal, and accessibility rules into the generation workflow.