PromptForge grades and improves prompts without calling a language model at all. A rule-based local engine scores what you type across seven dimensions — clarity, specificity, structure, context, output specification, constraints and role — and updates a weighted 0-to-100 score and letter grade as you write. Click Optimize and it rewrites the prompt at one of three strengths, showing a side-by-side before-and-after diff and a list of exactly what changed. You can steer the result towards a known framework (RTF, CRISPE, RISEN, chain-of-thought or few-shot) and format it for a specific target model — XML tags for Claude, Markdown for GPT — along with tone, audience and length controls. Because it is deterministic and entirely local, results are reproducible and nothing you write ever leaves your computer, which is the point for anyone drafting prompts that contain confidential material.
What it does
Live scoring across seven dimensions with a weighted 0-100 composite and letter grade
One-click optimisation at Light, Balanced or Aggressive strength with a before-and-after diff and change log
Framework targeting: RTF, CRISPE, RISEN, chain-of-thought, few-shot, or automatic
Formats output for Claude (XML tags), GPT (Markdown), Gemini or generic
Tone, audience and length controls
Interactive suggestion list with one-click fixes that re-score instantly
Template gallery with worked before-and-after examples
Personal saved-prompt library with replayable optimisation history
No API key and no network calls — deterministic, reproducible and completely private
Windows installer (.exe) — installs in a couple of clicks, no extra runtimes needed
Platform:
Windows 10/11 (64-bit)
Version:
0.1.0
Download size:
78 MB
Requires:
Runs fully offline. No account, no subscription, no internet connection needed. Not code-signed yet, so SmartScreen may warn on first run - choose More info, then Run anyway.
Licence:
Free forever for writing, live scoring, suggestions, saving to your library and importing or exporting it. A licence key unlocks running the optimiser and exporting the optimised result.
