Kept byte for byte
Code blocks, inline code, links, e-mail addresses, file paths, function and variable names, flags, numbers, versions, quotes and ALL-CAPS names.
When you write to ChatGPT, Claude, Lovable or any other AI, Typo first fixes your typos, then trims the filler: the pleases, the basicallys, the wordy phrases. Code, links, numbers and quotes stay exactly as you wrote them, and a single “not” is never lost.
This runs the same rules as the app, here in your browser. Nothing you paste is sent anywhere.
Typo already fixes your spelling everywhere. The token language is the second language: it only wakes up where it belongs, and always after the spelling is right.
You type “pleaze explan in ordr to”. The spelling engine makes it “please explain in order to” before anything else happens.
On the corrected text it removes “please” and turns “in order to” into “to”, at the level you chose, one finished word at a time.
Backspace right after a change puts back exactly what you typed. Ctrl+Alt+T shortens a whole prompt in one go, Ctrl+Alt+Z brings the original back.
In Auto mode the token language is active only in AI chat apps (ChatGPT, Claude and similar desktop apps) and on AI chat sites in your browser. Everywhere else, Typo only fixes spelling. It never acts in password fields, payment pages, terminals or code editors. Prefer it everywhere? Choose Always, or switch it off.
Level 3 is the default. Every level includes the ones below it. At every level, the same promises hold.
Every rule is checked before it is kept. If a check fails, that rule is dropped for that text.
Code blocks, inline code, links, e-mail addresses, file paths, function and variable names, flags, numbers, versions, quotes and ALL-CAPS names.
“not”, “never”, “without”, “inte”, “aldrig” and strength words like “only”, “must”, “always” and “all” are never added or removed.
Every original Typo changes is saved first in an encrypted file on your computer for 30 days. If it cannot be saved, your text is not changed. Copy any original back from Settings.
A plain log shows which rule ran, where and how many tokens it saved. It never contains what you wrote.
Everything runs on your computer. What you write is not uploaded, and this page runs the demo in your browser only.
Rules for both languages. Typo picks the language from what you write.
Typo recognises most of these tools and switches on by itself. Honest note: shorter text is not a promise of lower bills. Each provider counts usage in its own way.
The app and chatgpt.com. A clear goal, format and limits help more than clever abbreviations, so Typo trims filler and leaves your instructions intact. Projects are the better place for text you reuse.
OpenAI: prompt engineering best practicesThe desktop app and claude.ai. Typo trims filler and repeats. Usage in Claude also depends on conversation length, files and the model, so keep recurring background in a Project and point to precise files in coding work.
Claude: usage limit best practiceslovable.dev project chat. Lovable bills by the work a request causes, not by its length, so a shorter prompt is not automatically cheaper. Typo keeps scope, acceptance criteria and file names exactly as you wrote them and only removes filler. From Typo 0.12 lovable.dev is on the Auto list.
Lovable: credits and usageGemini, Copilot, Perplexity, Mistral Le Chat, Grok, DeepSeek, Poe, Kimi, Qwen and other AI sites are recognised too. If a site is not recognised, choose Always or use Ctrl+Alt+T.
Token language FAQSources reviewed 8 November 2026. Provider terms change; check each provider for current limits and prices.
Token counts here are estimates (about 4 characters per token), not the provider’s own count. Characters saved are measured exactly. Whether a shorter prompt lowers your bill depends on the provider, the model and how your plan counts usage. We do not claim savings we cannot measure.
Two places to learn what the token language does for the AI tools you use.
Pick ChatGPT, Claude, Lovable, Gemini or another tool and a level. See the shorter prompt, every change, estimated tokens and advice for that tool.
Open the Platform LabShort, honest guides to clearer prompts, by AI tool and by role.
No. It is a second language inside Typo, with its own page because many people want it on its own. One install gives you spelling correction and the token language.
It is built not to. Code, links, numbers, quotes and names are kept exactly, and words like not, never, only and must are never added or removed. Every rule is checked and dropped if a check fails. Higher levels read more clipped, so check the result at level 8 to 10 or turn on safe mode.
In Auto mode, only in AI chat apps and on AI chat sites. Typo reads the app name, and in browsers the page address, with the tab title as a fallback. You can also choose Always or Off.
Yes, first. Typo corrects the typo, then the token language works on the corrected text.
Yes. Press Backspace right after a change while typing. For a whole prompt, press Ctrl+Alt+Z. With safe mode on, every original is also kept for 30 days.
Only real spelling fixes count. A pure token change does not.
It makes your text shorter, which can help where usage depends on length. It is not a guarantee: providers count usage differently and some bill by the work done. We show estimates, not promises.
No. The token language runs on your computer. The demo on this page runs in your browser only.
Windows 10 and 11 and macOS 12 or later. Typing mode works on both. The whole-prompt shortcut is on Windows first.
Download Typo free: EUR 0 with 500 fixes a month for the first 3 months, or Typo Unlimited at EUR 5.90 a month. The token language is included in every plan and is on by default in AI tools.
Download Typo free