Paste two versions of a prompt and see exactly what changed, word by word, plus structural observations on what each edit does. Git diff for prompts. Runs in your browser.
Most prompt work is editing the same prompt over and over, but it’s hard to see what actually changed
between two versions and why the output moved. This shows a real word-level diff (like a code
diff) and then reads the structural observations: did you add an output format, remove a vague
word, resolve a conflicting instruction, change the token count? It uses the real o200k_base
tokenizer for the counts and the same checks as the Prompt Linter.
These are observations, not a verdict. The tool won’t tell you which prompt is “better”, only a real model on your real inputs can, and the honest move is to test both. What it does is surface the kinds of changes that usually matter, so you know what you changed. Nothing is uploaded.
No, and that’s deliberate. “Better” depends on your task and real outputs. It reports what changed (added format, removed vagueness, token delta) so you can judge, and run an A/B test yourself.
The diff is word-level: matching runs stay neutral, removed words are struck through, added words are highlighted. The observations look at structure, format, role, vague terms, conflicts, examples, and length.
No. Diffing and analysis run entirely in your browser, so both prompts stay on your device.
Lint a single prompt with the Prompt Linter, count tokens with the Token Counter, or browse AI Explained.