Help

Basic usage

1. Paste or type text in Input.
2. Choose Input format (or leave it on Auto to detect at transform time).
3. Choose Output format, then press Transform.
4. Use Copy output or Download to take the result with you.

For CSV output, open Separator to set the field delimiter. Normalize invisible characters cleans unusual spaces and line endings before processing — useful for text copied from web pages or PDFs.

Pasting from an AI chat

Dragging to select text in a chat window often drops headings, bullet points, and code blocks before you paste. Use the chat app's Copy button on the assistant message instead — it captures the full reply consistently.

• ChatGPT: use the copy icon at the bottom of the assistant turn.
• Google Gemini: use the message-level copy action in the web app.
• Anthropic Claude: use the copy icon on the assistant message — more reliable than Ctrl+C after selecting for long replies.
• Microsoft Copilot: use the copy control on the assistant response.

Avoid selecting both your message and the assistant reply in one sweep — copy the assistant turn only.

Plain text formatted (AI-ready)

Markdown works well for LLM output and longer documents — but in system prompts it works against you. Every token spent on ## or **bold** is a token you can't spend on an actual instruction. Plain text formatted keeps the structure using conventions LLMs already understand — ALL CAPS for headings, *asterisks* for emphasis — so your entire token budget goes to meaning.

## Role
You are a **support** assistant.

### Rules
- Never speculate

ROLE
You are a *support* assistant.

  Rules
- Never speculate

How the token count is calculated

The ~N tokens figure shown below each text area is a rough estimate — not an exact count. It uses the rule of thumb that one token corresponds to roughly four characters in typical English text. The formula is simply characters ÷ 4, rounded to the nearest integer.

The actual token count you will be charged for depends on the model's tokenizer, which is not publicly available in a form that can run in the browser. Different models — GPT-4o, Claude, Gemini — use different tokenizers, and the same text can produce slightly different counts across them. As a rule:

• English prose lands close to the 4-chars-per-token estimate.
• Code and punctuation-heavy text tends to be more tokens (more short tokens, symbols tokenized individually).
• Languages with longer words (German, Finnish) tend to be fewer tokens per character.
• Non-Latin scripts (Chinese, Arabic, Japanese) often use more tokens per character than Latin text.

The tilde (~) prefix is intentional — treat the number as an order-of-magnitude signal, not a billing figure. For exact counts, use the tokenizer tool provided by your model vendor.