Editor

Paste an essay or upload a PDF, then choose what kind of help you need — feedback, rewrites, or generation. The prompts are engineered, you don't write a thing.

Why it exists

Researchers and writers already use Claude or ChatGPT to critique their drafts. The friction: the prompts that produce useful feedback are long, specific, and tedious. “Find logical fallacies” as a one-liner gets you generic high-school-debate categories. The prompt that gets named fallacies anchored to specific sentences with repair suggestions is 200 words, and you don't want to type it every time.

Kvire ships those prompts. You click an action; you get the output.

Three categories

• Feedback — read the draft and tell you what's strong, what's weak, what to fix. AI does not produce new text. Includes: general feedback, logical fallacies, weak arguments, best counter-arguments, concrete suggestions, clarity check.
• Rewrite — produce alternative versions. Includes: new opening (3 angles), sharper thesis (3 versions), new conclusion, tighten 30%, more academic, more accessible.
• Generate — derive new content from the draft. Includes: title ideas, one-line summary, academic abstract, Twitter/X thread.

Inline rewrite

Highlight any passage in your draft and a ✨ Rewrite button appears. Click it and pick a style preset (tighten, clarify, more formal, strengthen, soften, expand, active voice…) or type a custom instruction. Kvire produces a rewrite that fits seamlessly back into the surrounding text — it knows what comes immediately before and after the selection so tense, register, and continuity match. Replace the passage in your draft with one click.

How it works

1. Open /editor.
2. Paste text (≥150 words) or upload a PDF (≤25 MB). Stored on your account.
3. Pick a category, then pick an action — feedback streams in.
4. Run as many actions as you want. Cached so re-opening costs nothing.
5. For sentence-level edits, highlight a passage and use inline rewrite.

Cost

The Editor uses Anthropic prompt caching: the document is sent once with cache_control: ephemeral, then each action runs against the cached context. Cached input tokens cost ~10% of full input, so running half a dozen actions on a 6000-word paper costs roughly $0.05–$0.07 in API credits instead of $0.30+ without caching.

What it's not

• Not a copy-editor. Sentence-level grammar isn't the goal — Grammarly is good at that.
• Not editorial. Output is AI-generated; treat it as a peer-review pass, not a final judgment.
• Not for scanned PDFs. Image-only PDFs can't be extracted.

Engineered prompts, not vibes

Every action in the Editor is backed by a prompt that specifies role, output structure, anti-invention safeguards, and explicit voice. The fallacy detector requires named fallacies anchored to quoted text with repair suggestions — not a list of generic debate-class categories. The opening generator produces three distinctly different angles (provocative / concrete / stakes-first) and is forbidden from using dead phrases like “In recent years.” The tightening pass cuts hedge words and throat-clearing on a per-pattern list. The product is the prompts, not the wrapper.