Prompt Compressor
Paste a bloated prompt, get a leaner one. A rule-based pass strips filler words and verbose phrasing, then shows exactly how many tokens — and dollars — you saved. 100% client-side: your text never leaves your browser.
Conservative: only trims formatting, filler words and wordy phrases. Sentence structure stays untouched.
Prices verified 2026-10-01 from public sources. Confirm on the provider's pricing page before budgeting real spend.
Tokens saved
What changed
Red strikethrough = removed, green = replacement text.
Compressed prompt
An honest note before you trust the output
This tool is a fixed set of find-and-replace rules — no AI, no language model, no understanding of your intent. Conservative mode is safe for almost any prompt: it only removes filler and wordy phrasing. Aggressive mode rewrites sentence openers and can shift tone or emphasis. A rule engine cannot know which words carry meaning in your specific task, so always re-read the compressed prompt before sending it to a paid API — especially for system prompts that run thousands of times.
Why compress prompts at all
In a chat app the meter is invisible. Move the same prompt to the API — an agent loop, a batch job, a RAG pipeline — and every token has a price tag. Bloated system prompts are the classic leak: "please note that it is very important to" costs tokens on every single call while adding zero information. Trimming 15–25% off a system prompt that runs 10,000 times a month is real money, and it often makes the model follow instructions better, because the signal-to-noise ratio goes up.
How the estimates work
Token counts use the same heuristic as our Token Counter: words × a per-family factor plus CJK characters × a CJK factor (GPT 1.33/1.5, Claude 1.35/1.6, Grok 1.33/1.5). Real tokenizers differ, so treat the numbers as estimates within ~20% for plain English. Cost uses the selected model's public input price per 1M tokens, verified 2026-10-01.