People burn money and run out of tokens mid-session because they use the wrong AI model for the prompt — and prompt gateways just take that decision away without teaching anyone anything. Fidget is an AI coach that reads your prompt before you send it, rewrites it tighter to save tokens, and recommends the right model and effort — explaining why, so you save cost, keep the decision, and get better at AI every time you type.
Never run out of tokens during a session — the right model makes your quota last.
Rewrite your prompts to save tokens — tighter prompts, same answers, less cost.
Enable your team to write better prompts and choose the right model.
“Prove that the square root of 2 is irrational, showing each step.”
→ Sonnet · Medium effort — a proof needs real reasoning, but not the priciest model. ~9 tokens saved with a tighter rewrite.
Fidget reads the prompt you're about to send in ChatGPT, Claude or Gemini, rewrites it tighter to save tokens, and tells you which model and effort to use — and why. Other gateways take the decision away from you; Fidget gives it back, so you write better prompts, learn which model fits, and stop paying for more model than your prompt needs.
Install the extension — it lives in ChatGPT, Claude and Gemini on the web.
Type your prompt as normal. Fidget analyses it instantly, right where you type.
Pick the recommendation, copy the rewrite, send. The right model, the right effort, fewer tokens.
One small panel, bottom-right, before you hit send — and it pays for itself.
Instant vs Medium vs High, Haiku vs Sonnet vs Opus, Flash vs Pro — matched to what you actually asked, so a simple question never burns your best model.
Low, Medium, High or Max reasoning, matched to the task. When easy questions stop eating your quota, your session lasts until you're done — not until the cap hits.
A tighter rewrite of your prompt, with the tokens you save. Copy it in one click — same answer, less spend.
Most companies try to control AI by restricting it — which models people can use, how much they can spend, who gets access. That caps your team's growth, and the cost worry never goes away.
Fidget flips it. It teaches every person on your team to use AI well — the right model, the right effort, the tighter prompt — so they get more done for less. You empower your people to use AI to the maximum, instead of limiting them and watching the bill. A prompt gateway does the opposite: it picks for you and moves on, so nobody learns anything. Fidget shows you the why, every time, so the team gets better with every prompt.
Everything is free while we're early — enjoy the full product now. Paid tiers are coming later.
Get it from the Chrome Web Store — one click to install, and it updates itself automatically.
Click Add to Chrome on the Store page.
Open ChatGPT, Claude or Gemini, click the prompt box and start typing. No key needed — the backend does the AI.
When the bubble appears, agree to the consent prompt once — that's the only setup.
Prefer to install manually? Download the .zip, unzip it, then open chrome://extensions, turn on Developer mode and click Load unpacked. Manual installs don't auto-update.
For the AI recommendation, your prompt is sent to Fidget's backend and Google's Gemini API, and processed there in the cloud — there is no on-device AI. Your prompt text is not stored; the recommendation is cached for up to 30 days under a one-way hash of your prompt so repeats come back instantly. Separately, anonymized prompts (emails, API keys, phone numbers, and card numbers stripped out first) are shared to improve Fidget — that's on by default, and you can turn it off any time in the extension popup.
No. There is no key to add — everything runs on Fidget's backend, free while in beta. Just install the extension and start typing.
Because even on a paid plan you hit limits and wait for your session to reset. Fidget makes the plan you already pay for go further — the right model for each prompt, so your quota lasts and you get pro-level answers without upgrading.
They're estimates, not a bill. The token count is roughly 4 characters per token, the reply length is estimated from the task and effort, and model prices change. They're good enough to compare models and spot an expensive prompt before you send it — not exact to the cent.
Any Chromium browser — Chrome, Edge, Brave, or Arc — version 138 or newer. It doesn't work in Firefox or Safari.
Not yet. Browser extensions can't reach the native iOS/Android ChatGPT, Claude or Gemini apps. The extension works on the web versions, on desktop and on mobile browsers that support extensions.