AI Prompt Improver
Rewrite weak one-line prompts into structured, model-ready prompts. Analyzes clarity, specificity, context, constraints, format, and examples. Adds role framing, explicit task, context slots, constraints, and output format. Scores 0-100 with tips. Target-model presets (GPT-4, Claude, Gemini, local). Variable extraction + reusable template. Pure-JS engine — optional BYO-key LLM. 100% client-side, nothing uploaded.
About AI Prompt Improver
Rewrite weak one-line prompts into structured, model-ready prompts. Analyzes clarity, specificity, context, constraints, format, and examples. Adds role framing, explicit task, context slots, constraints, and output format. Scores 0-100 with tips. Target-model presets (GPT-4, Claude, Gemini, local). Variable extraction + reusable template. Pure-JS engine — optional BYO-key LLM. 100% client-side, nothing uploaded. Everything runs locally in your browser — your data never leaves your device.
How to use
- Enter your input in the tool above.
- Adjust any options to your preference.
- Use the Copy or Download buttons to save the result.
- Everything happens locally — your data never leaves your browser.
FAQ
How does the AI Prompt Improver work?
Paste your rough prompt and the analyzer scores it across six dimensions — clarity, specificity, context, constraints, format, and examples — yielding a 0-100 score with per-dimension breakdowns. The rewriter then adds missing components: a role framing line, an explicit task statement, context slots, constraints (length, tone, things to avoid), a desired output format, and a few-shot scaffolding block. You get an improved prompt, a diff of what changed, and a per-change explanation. By default everything runs on-device with deterministic rules.
What target models are supported and what changes between them?
Four target presets: GPT-4-class, Claude, Gemini, and Small Local. Each preset tweaks phrasing conventions — for example, Claude favors explicit XML-style section tags (<context>, <task>) and a thinking-friendly structure, GPT-4-class favors clear numbered steps and JSON output schemas, Gemini favors concise role + bullet constraints, and Small Local favors shorter prompts with explicit format examples to compensate for limited context handling. The rewriter applies the preset's conventions automatically.
What is the variable extraction / template mode?
Variable mode scans the improved prompt for concrete values (names, numbers, technologies, file paths) and turns them into {{double_brace}} placeholders, producing a reusable template. You can fill the placeholders in any future run by typing values into the template form, without rewriting the prompt. Templates persist in localStorage and export as JSON.
What extra features does this tool have compared to others?
(1) Six-dimension prompt analyzer (clarity, specificity, context, constraints, format, examples). (2) 0-100 score with per-dimension breakdown and tips. (3) Rule-based rewriter adding role, task, context, constraints, format, and few-shot scaffolding. (4) Four target-model presets. (5) Diff view (added/removed lines) with explanations. (6) Variable extraction and reusable template mode. (7) Token-budget estimator (rough). (8) Negative-constraint builder ('avoid X'). (9) Preset use-cases (coding, writing, image-gen, analysis, agent-system-prompt). (10) Copy improved prompt, copy template, export JSON. (11) History (localStorage, last 20). (12) Shareable URL. (13) Optional BYO-key LLM enhancement. (14) Deterministic — same input always produces the same output.
Is my prompt sent anywhere?
No. All analysis, scoring, rewriting, and variable extraction run locally in your browser. Prompts never leave this device. The only network call is if you paste your own LLM API key and click 'Enhance with LLM' — that request goes directly from your browser to the LLM provider you choose.
Related tools
AI .htaccess Redirect Generator
Generate Apache .htaccess redirect and rewrite rules from bulk old→new URL pairs or plain-language intent. Supports 301 permanent, 302 temporary, 307, RedirectMatch, RewriteRule with regex capture groups, www/non-www canonicalization, HTTP→HTTPS, hotlink protection, custom error pages, and IP allow/block. Per-rule explanations, sample-URL tester emulating mod_rewrite, and redirect-loop detection. Pure-JS engine — optional BYO-key LLM. 100% client-side, no upload.
AI Alt Text Generator for Images
Generate WCAG-friendly alt text for images. On-device metadata extractor (dimensions, format, dominant color, filename → suggest), context-aware descriptive templates, decorative vs informative vs complex classifier, length meter, 'don't start with image of' linter, optional SEO keyword weaving, 100+ language templates, batch CSV/JSON export, optional BYO-key LLM caption enhancement. 100% client-side — images never uploaded.
AI Analogies Generator
Explain any concept with tailored analogies across 10+ domains (cooking, sports, nature, technology, music, business, science, everyday life, kids, vehicles). Pure-JS analogy template engine with metaphor / simile / story styles, complexity levels, 'where it breaks down' honesty notes, explainer expansion, concept keyword extractor, audience presets, optional BYO-key LLM enhancement. 100% client-side — nothing uploaded.