ContextMeter

Know when your AI chat is running out of room

ContextMeter shows how full the context window of your ChatGPT or Claude chat is, tells you before answers start slipping, and carries the chat into a fresh one with a single click.

Add to Chrome: soon Add to Firefox: soon
  • Free, no account
  • Nothing leaves your browser
  • Chrome, Edge and Firefox

Store links will appear here once the extension is approved. In the meantime, try the free token counter.

Try it: drag the slider
0%

~0 of 16K tokens

Plenty of room

In short

What is ContextMeter?

ContextMeter is a free browser extension for Chrome, Edge and Firefox that shows how full the context window of a ChatGPT or Claude chat is. It counts the tokens in the conversation, shows the percent of the window used with a gauge that turns amber and red, detects your ChatGPT plan, and offers a one-click handoff that moves the conversation into a fresh chat with a structured summary. It runs entirely on your device: no account, no analytics, no servers, and message text is never stored. It is independent and not affiliated with OpenAI or Anthropic.

New to this? Read what the ChatGPT context window is, when to start a new chat, or compare token counter extensions.

The problem

Long chats quietly get worse

Every reply is written from your whole conversation. As it grows, the model has more to juggle, older details fall away, and the chat starts forgetting what you decided. Neither site tells you how close you are.

It forgets

Early requirements and decisions drop out of view, and the answers stop matching what you asked for.

It repeats itself

Fixes you already rejected come back, and corrections stop sticking.

You find out late

There is no gauge in the chat, so you usually notice only after an hour of worse answers.

What you get

A meter in the corner, and a way out when it fills

Below is the real extension on a sample conversation, not a mock-up.

See how full the chat is, at a glance

A small percent pill sits beside the chat and turns amber, then red. Open it for tokens used, about how many more messages fit, and a strip showing which messages take the most space.

On ChatGPT it detects your plan (Free, Plus, Pro) and the model from the page, so the number is measured against the right window. You can override any limit.

ContextMeter's panel open next to a ChatGPT conversation, showing a gauge, tokens used, the plan and a message map
On a sample ChatGPT chat. Plan and model are read from the page.

Move to a fresh chat without losing the thread

When the chat is nearly full, Handoff asks the AI for a structured summary: goal, decisions, constraints, key facts, next steps. Then it opens a fresh ChatGPT or Claude chat with that summary pre-filled.

You press Send yourself, and the new chat is asked to restate the key facts so you can see it understood.

Get the handoff prompt to use by hand

ContextMeter showing a captured handoff summary with buttons to open a new ChatGPT or Claude chat
A captured summary, ready to open in a new chat.

Counts the whole chat, even long ones

ChatGPT and Claude open long chats showing only the newest messages and load the rest as you scroll. So a naive counter under-reports. With your permission, ContextMeter reads the full chat quietly while you are idle, puts you back where you were, and remembers the result.

It asks first, shows a card while it works, and has a Stop button and an off switch.

ContextMeter panel in dark mode on a Claude conversation, with a scan full chat button and an automatic scan option
On Claude, in dark mode.
Draft previewSee what an unsent paste will cost before you send it.
Honest numbersLabeled as estimates of visible messages, with a lower bound when part of the chat is not loaded.
PrivateNo account, no analytics, no servers. Message text is never stored.
Keyboard friendlyAlt+Shift+M opens the panel. Passes automated accessibility checks.

How it works

Three steps, no setup

Install it, open a chat, and keep an eye on the corner.

  1. Open a chat

    On ChatGPT or Claude, a small percent pill appears next to the chat.

  2. Check the details

    Click it for tokens used, room left and the biggest messages. Press Alt+Shift+M if you prefer the keyboard.

  3. Hand off when it is red

    Start a fresh chat with a summary of everything that mattered, instead of starting from zero.

Handoff

Start fresh without starting over

When a chat is full, the usual advice is "start a new one", which throws away everything the model learned about your task. Handoff keeps the useful part.

  • Asks the AI for a structured summary: goal, decisions, constraints, key facts and next steps.
  • Opens a fresh chat with it pre-filled, on ChatGPT or Claude.
  • Includes a self-check so you can see the new chat really understood. You always press Send.

Get the handoff prompt to use by hand

Privacy

Built so you can trust it

An extension that reads your chats should earn trust. This one is deliberately boring about it.

What it does

  • Counts the messages shown on the page, in your browser.
  • Remembers message ids and token counts so chats open instantly.
  • Runs only on chatgpt.com and claude.ai.

What it never does

  • Send your chats, counts or anything else to a server.
  • Store message text.
  • Require an account, run analytics or show ads.

Read the full privacy policy

Context limits

How big is your window?

It depends on the model and plan, and the numbers change often. These are the defaults the extension starts with; every one is editable.

16KChatGPT Free
32KChatGPT Plus
128KChatGPT Pro
200KClaude standard

Sources disagree and OpenAI and Anthropic change these. See the full table with sources.

Questions

Frequently asked

What is a context window?

It is the amount of text, measured in tokens, that an AI model can consider at once. Your prompt, the whole chat so far and the reply all have to fit. When a chat gets close to the limit, older parts are dropped or matter less and answers get worse.

Is the count exact?

No. ContextMeter counts the message text visible on the page. The model's real context also includes things the page does not show, like system prompts, memory, tool output, hidden reasoning and attachments, so treat the number as a guide. ChatGPT chats use a real tokenizer; Claude uses an estimate because Anthropic does not publish an in-browser tokenizer.

Does it read or send my chats?

It reads the visible messages in your browser to count them. Nothing is sent anywhere, there are no accounts or analytics, and message text is never stored; only message ids and token counts are remembered so a chat you have scanned loads instantly next time.

Why does it scroll or reload my chat?

ChatGPT and Claude load long chats in pieces, so the only way to count a whole chat is to load it. If you allow it, ContextMeter does this quietly when you are idle, puts you back exactly where you were, and remembers the result. It asks first and has an off switch in the panel and in settings.

Which sites and browsers does it support?

ChatGPT and Claude on Chrome, Edge and Firefox. Gemini and DeepSeek are planned.

Is it free?

Yes. There is no paywall and no account.

Is it affiliated with OpenAI or Anthropic?

No. It is an independent tool.

Questions, bugs or ideas?

Write to us at support@contextmeter.app. We read every message.

Contact us

Stop guessing how full your chat is

Free, private and small. Install it once and glance at the corner.