metacogni.ai
MetaCogni for Chrome · early access

thinkingaboutthinking.

MetaCogni is the browser layer for AI subscriptions. It reads the prompt before you send, predicts output tokens, diagnoses weak instructions, and shows the limit that matters for Claude, ChatGPT, or Gemini — because flat subscriptions still cost you in time, quality, and caps.

claude.ai · new chat
>
predicted: 1.2k tokens intent: audit Claude: 9% session burn Brain says: add success criteria
Add to ChromeWatch it think

Free during beta · on-device prompt analysis · no prompt logging

Claude Opus 4.7
claude.ai
Audit this backend auth change before I burn Opus time.
MetaCogni catches the expensive part first: long thread, audit intent, cold cache, missing success criteria
Claude Opus 4.7
claude.ai
Session burn14%
Session
14%
Cache
warm
0.27 API-equivalent · 4h 41m left
code_gen+820 tok est.

A small extension that thinks about your thinking.

Four browser-side checks run before you hit send: intent, output size, limit pressure, and prompt quality. Your prompt stays on the page you typed it on.

metacognition · n.

Shows the cost before the thought leaves.

Every prompt has a likely shape: short factual answer, long explanation, code fix, audit, comparison, or creative draft. MetaCogni estimates that shape before you commit the message.

intent classifier

Classifies intent without sending the prompt away.

A local intent classifier recognizes factual, yes/no, math, code-gen, code-fix, explain, creative, list, comparison, conversational, and general asks. The browser does the first read.

limit-aware forecast

Tracks the meter that actually bites.

Claude has 5-hour burn, weekly model quotas, and extra-usage billing. ChatGPT has rolling message caps. Gemini has daily Pro limits and long-context quality risk. MetaCogni separates those instead of pretending every plan is dollars.

prompt rewriter

Diagnoses the prompt, then prescribes the nudge.

Brain Says recommends the best directive to append. Could Be Better flags missing format, scope, success criteria, or context risk. The goal is fewer wasted tokens and better answers.

Live preview · animated

See your session burn in real time.

The overlay watches the prompt box, estimates the reply before send, and updates the relevant meter: Claude session burn and weekly quotas, ChatGPT message caps, or Gemini daily/context pressure.

C
Claude
Opus 4.7 · long thread
claude.ai
you
Fix this React hydration bug. I pasted the component and server log above.
code_fix~420 tok+4%
MetaCogni overlaylive
Session burn8%
Claude Pro · 5h session2h 41m left
Est. output
~420 tok
Messages
1 sent · 0 received
Latest intent
code_fix
Suggestion

Cache is warm. Keep going — context is cheap right now.

Lives where you already do

One extension. Three different limit systems.

Claude burns sessions, weekly quotas, and sometimes extra-usage dollars. ChatGPT spends rolling message caps. Gemini spends daily Pro queries and long-context quality. MetaCogni keeps those meters separate.

Capability
claude.ai
claude.ai
ChatGPT
chatgpt.com
Gemini
gemini.google
On-device · every platform
11-intent classifier
Fast local prompt intent detection before send
Output-token prediction
Short vs verbose reply estimate before you spend the turn
Concise-mode rewrites
Suggested directives for factual / list / code-fix / yes-no prompts
Model-fit hint
Flags when the big model is unnecessary for the prompt shape
Brain panel · diagnose & prescribe
Brain Says recommends a nudge; Could Be Better explains the weak spot
Platform-specific tracking
Extra-usage billed delta
When Claude extra usage is on, show the billed currency delta
—
—
5-hour session burn meter
Claude's primary throttle window
—
—
Weekly Opus / Sonnet quota
Separate seven-day buckets when Claude exposes them
soon
—
—
Cache-warm vs cold cost
Shows when repeated Claude context is likely warm or cold
soon
—
—
GPT-4o cap countdown
Show how close the rolling message window is to the cap
—
soon
—
o3 / o4-mini routing hint
Reasoning intents vs light asks get different suggestions
—
soon
—
Daily Pro query counter
Track Pro-query pressure when Gemini exposes the count
—
—
soon
Long-context drift warning
Prompt to trim and re-anchor when long threads get noisy
—
—
soon

= next adapter layer · current beta already runs local prompt analysis

11
prompt intents recognized
0
prompts sent to a server
3
AI platforms supported
<5ms
on-device classification

Knows your plan. Tracks the limit that bites.

Each subscription fails differently. MetaCogni shows the meter you should care about right now: session burn, weekly quota, rolling messages, daily queries, or context risk.

Anthropic
Claude Pro
$20/mosubscription
5h session + weekly caps
best tracked on Claude today
Anthropic
Claude Max 5×
$100/mosubscription
larger 5h + weekly pool
higher ceiling, same meters
most tracked
Anthropic
Claude Max 20×
$200/mosubscription
largest Claude pool
weekly Opus matters most
OpenAI
ChatGPT Plus
$20/mosubscription
rolling message caps
model choice changes pressure
OpenAI
ChatGPT Pro
$200/mosubscription
higher GPT + reasoning access
soft caps still matter
Google
Gemini Advanced
$20/mosubscription
daily Pro-query pressure
long context needs hygiene

Limits are approximate and provider-controlled. MetaCogni uses public limits where necessary and live platform signals where available.

ready when you are

Know before you send.
Then spend the good model on purpose.

MetaCogni runs across Claude, ChatGPT, and Gemini. No API keys, no signup, no prompt logging — just a quiet browser layer that estimates the next reply and tells you how to make the prompt cheaper or better.

Download extension (.zip)View on GitHub
Sideload in 3 steps~30 seconds
  1. 1Unzip the download. The folder you want is browser-extension/.
  2. 2Open chrome://extensions and toggle Developer mode on.
  3. 3Click Load unpacked and pick that folder. Visit Claude, ChatGPT, or Gemini — the overlay appears on supported chat pages.
Or try the BYOK workspace →Chrome 116+ · Edge · Brave
Local-first
Open source
MIT licensed