Repository navigation
[Feature Request] Add format: 'html' option to support HTML input #32
Description
Summary
While testing the Proofreader API (Origin Trial), I found that the model aggressively removes HTML tags from the input, treating them as noise or errors to be cleaned.
Steps to Reproduce
- Open the official Proofreader API Playground/Demo (or run locally).
- Input the following text containing an HTML tag and a typo:
<div>The div is not closd - Run the proofread function.
Actual Result
The model returns: "The div is not closed."
The <div> tag is completely removed from the output string.
Expected Result
The model should correct the typo but preserve the HTML structure:
<div>The div is not closed.
Impact
This behavior makes the API unsuitable for use cases involving technical documentation, code editors, or rich-text environments where markup must be preserved.
Activity
tomayac commentedon Nov 21, 2025
What you're actually asking for is for a new input type, HTML. I had filed similar feedback in webmachinelearning/translation-api#60 for the Translator API, where HTML happens to be supported surprisingly well, but more as a lucky coincidence and not because of official support.
vero-code commentedon Nov 21, 2025
Thanks for referencing this in issue #60 — that helps clarify the broader discussion.
Having a standardized
format: 'html'option would definitely make sense. For workflows involving rich-text editors or CMSs, preserving markup isn’t optional for these use cases.tomayac commentedon Nov 21, 2025
Could I ask you to update the Issue title to reflect the
format: 'html'feature request? Thank you![-][Bug] Destructive HTML handling: Model strips tags from input[/-][+][Feature Request] Add `format: 'html'` option to support HTML input[/+]vero-code commentedon Nov 21, 2025
Done! I've updated the title as requested.