Green · Clear & direct
“That is incorrect. Check the math.”
A firm request is still a good request. No obvious harsh phrasing detected.
LOCAL FIRST. HUMAN FRIENDLY.
Less heat.
Better prompts.
For the moments your AI needs clearer instructions—and your keyboard needs a minute.
A tiny tone coach for Claude and ChatGPT. Catch the harsh bits. Keep your point.
Free beta. No account. No compliment sandwiches.
On-device checksNo trackingYou control Send
YOUR NEXT MESSAGE
A very patient autocomplete LIVE DEMO
Type something. Watch the bar. Click it for a little perspective.
Fits right under your message.
No new workflow. Just a second thought.
ClaudeChatGPTGemini, later.
01 / READ THE ROOM. EVEN THE VIRTUAL ONE.
Three colors.
One useful nudge.
It checks the wording, not your worth as a human. Directness is welcome. Flattery is entirely optional.
Amber · Getting sharp
“You never listen.”
The keyboard is getting warm. Point to the requirement that was missed.
Red · Sounds harsh
“You’re a useless idiot.”
Keep the feedback. Lose the flamethrower. Get a targeted edit you can review and copy.
Gray means not rated. English-first heuristics can miss context; no color predicts policy compliance or account safety.
02 / PRIVATE BY DEFAULT
Your draft.
Your business.
Local checks run on your device. No accounts, telemetry, ads, or saved drafts. The extension reads the current composer—not your chat history.
Nothing is automatically rewritten or sent. Suggestions are yours to review, copy, ignore, or argue with. It’s your keyboard.
Read the privacy details ↗OPTIONAL UPGRADE · SAME SMALL EGO
Two tone detectives
are better for a debate.
Ask Jev for a contextual second opinion. See where local rules and a decision model agree—or don’t.
Each check requires your confirmation. The draft goes to your configured relay and TypeSafe AI. Your Jev key stays on the server.
CLASSIFICATION, NOT A GENERATED REWRITE03 / SMALL DOWNLOAD. SMALLER EGO.
Meet your new
keyboard companion.
The early-access beta installs manually in desktop Chrome. A Chrome Web Store listing has not been published yet.
Claude and ChatGPT support is in beta. Site updates may affect the meter.
- Download and unzip.Keep the extracted extension folder somewhere permanent.
- Open Chrome’s extensions page.Type
chrome://extensionsin the address bar and enable Developer mode. - Choose “Load unpacked”.Select the extracted folder containing
manifest.json. - Reload Claude or ChatGPT.Start typing. Click the meter for feedback, or the extension icon for settings and the playground.
04 / A LITTLE CLARITY
Good questions.
Zero guilt trips.
Will this stop Claude from banning me?
No. Please.exe estimates wording, not account risk. Anthropic’s update, effective November 12, 2026, addresses sustained, needless abuse, explicitly excludes ordinary frustration and pushback, and says ending abusive conversations remains the primary enforcement mechanism. An ended conversation is different from an account ban.
Read Anthropic’s announcement · Checked October 11, 2026.
Do I have to say “please” to get a green bar?
No. “Fix this calculation” is perfectly fine. Specific instructions, constraints, and examples are more useful than adding politeness for its own sake. This is a tone coach, not a manners tax.
What about quotes, code, sarcasm, or another language?
Local rules skip quoted text and code, and treat technical phrases like “kill the process” normally. They are English-first and imperfect. Unsupported or insufficient text is not rated. Jev can offer another perspective, but neither method guarantees the correct interpretation.
What exactly does Jev do?
Jev is TypeSafe’s decision model. It classifies the draft and returns a structured verdict; it does not generate a full rewrite. Explanations are curated. It is optional, needs a configured relay and API key, and never runs automatically.
Can I use this at work?
Local mode does not transmit drafts. Follow your organization’s extension policy. Only enable Jev if your organization permits sending that draft to your relay operator and TypeSafe. Keep confidential material out of cloud checks.
Be direct. Be specific.
Maybe retire “useless idiot”.
A calmer prompt is still your prompt.