Your words, your voice - just properly written. One request, one reply.
The polished version will appear here.
Most writing tools fix your grammar and quietly replace you while they are at it - the message comes back smoother, longer and no longer sounding like anyone. This fixes the mechanics and leaves the voice alone.
The instruction to the model is narrow on purpose: correct what is wrong, leave what is merely yours. Short sentences stay short, a blunt sentence stays blunt, and your word choices survive unless they are actually errors. If you want something rephrased entirely, this is the wrong tool - that is a rewrite, not a polish.
Light fixes only outright errors and paragraph breaks. Balanced, the default, also tightens awkward phrasing. Firm restructures for clarity while keeping your register. Email lays the text out as an email body and does not add a subject line, greeting or sign-off you did not write; Chat message keeps it to one compact message; Bullets uses a short lead-in and bullet points; Prose never uses bullets. The nudge (only its first 500 characters are used) is treated as a style preference, never as permission to add facts.
With the site in English, the text comes back in the language you wrote it in, mixed-language writing included. With the site set to another language, the reply is written in that language, so English text pasted on the Spanish site comes back in Spanish. To keep a text in its own language, switch the site language (the globe at the top) to match it first.
It runs on a small, fast model (Gemini 2.5 Flash-Lite by default) with thinking turned off and the reply capped at 4,096 tokens, because correcting grammar does not need a reasoning budget. The estimate shown after each run counts about 4 characters per token: a 1,000-character email comes to roughly a hundredth of a cent, a full 12,000-character run to well under one cent.
Copy always takes the text as it appears under Text, whichever view is open: plain text that keeps any Markdown markers the model used, such as **bold** around a term you emphasised or "- " in front of list items. Preview only shows how those markers render. Editors and chat apps that read Markdown turn them into formatting; Outlook, Gmail and Word paste the characters as typed. For text going into an email client, pick Prose (no bullet points) and delete any leftover asterisks. A code fence wrapped around the whole reply is removed before you copy.
It should not. It corrects mechanics and structure; the words and the tone are meant to stay yours. Turn the editing strength down if you want it even lighter.
Around 12,000 characters per run - comfortably more than an email and less than a chapter.
Discuvr does not save it. The text is sent to Google's Gemini API to be polished and comes back in the same request. The usage record keeps only character counts, the format and strength you picked and whether you added a nudge. There is no chat history because there is no chat.
Yes. Each run calls a language model, so the tool needs a free sign-in, and each visitor can run it 30 times in 10 minutes.
They are Markdown markers for bold text or list items. Copy takes the plain text with the markers included; the Preview view shows the formatting but does not change what is copied. Delete them, or paste into an app that reads Markdown.
The reply is capped at 4,096 tokens. Google counts a token as about 4 characters, so a typical text up to the 12,000-character limit fits, but a reply that reaches the cap simply ends there and the page shows no warning. If the end is missing, polish the text in two parts.
Only up to a point, by design. It fixes errors and structure but keeps your register: a casual note stays casual, and the model is told not to add stock phrases such as "I hope this finds you well" or a greeting or sign-off you did not write. For a work email, pick Email as the format and Firm as the strength, which restructures for clarity, and put a style preference such as "shorter sentences" in the nudge. The nudge does not lift the rule against turning your text into corporate prose, so for a fully formal version, write the formal wording yourself and use this to clean it up.
Yes, with the site in English. The model is told to keep the original language, mixed-language writing included, so a Hinglish or Spanglish message comes back mixed the same way, with grammar, punctuation and structure fixed. With the site set to another language, the reply is written entirely in that language, so the same message on the Hindi site comes back in Hindi. Switch the globe at the top to English first if you want to keep the mix.