Hey Product Hunt! 👋
Founder here — thanks for checking out Cavebro AI Token Saver.
I built this after getting frustrated seeing the same advice repeated everywhere: "shorten your words to save tokens." Type "gnrt" instead of "generate." Use "u" instead of "you."
It sounds logical. But it doesn't work. The tokenizer used by OpenAI and Anthropic was trained on real English text, so common words like "generate" are already compressed into single tokens. "Gnrt" is an unknown pattern — it often costs more tokens than the word it replaces.
The actual waste is somewhere else entirely. It's the filler nobody questions:
"Can you please provide a comprehensive explanation of..."
"Thank you in advance, feel free to ask if you have any questions"
"I would like you to..."
These phrases survive in every prompt because they feel polite. But they mean nothing to a language model — and you're paying for every single token.
Cavebro removes them. One keystroke, any text field, no AI involved, nothing leaves your browser.
This is v1 — the pattern library is intentionally starting lean so I can grow it based on real usage. If you press ⌥Z on a prompt and it misses something obvious, I want to know. That's exactly how v2 gets built.
Would love any feedback from this community — you're exactly the kind of power users who'll stress test this properly.
Replies
Dorik