Save tokens with CONTEXA for Claude

Long session? Keep going.

Every message on claude.ai re-reads the whole thread. When it grows heavy, CONTEXA carries it into a new chat as a short brief: same session, fewer tokens. And after any reply, up to four next messages, already written. Nothing is sent until you click.

Add to Chrome View on GitHub

Free Chrome extension
No account or API key
Privacy-first

Nothing is sent for you

claude.ai · new chat

··· 38 earlier messages

Can you help me understand how to factor a quadratic when the number in front isn't one?

The trick is to split the middle term into two pieces, then factor by grouping.

The grouping step always trips me up — how do I know which two pieces to pick?

Look for two values whose product matches the outer terms and whose sum is the middle one, then rewrite and group.

That clicked. When would I complete the square instead of factoring?

Reach for it when you want the vertex form, or when the roots won't come out clean.

And that's where the quadratic formula actually comes from?

Exactly — the formula is completing the square done once, in general, for any quadratic.

Could we move on to how derivatives connect to all of this?

Brief ready

A fresh chat, with your thread carried over

Written in the thread's own words — nothing sent yet

CONTEXA ≈ 689k tokens off every message

This thread is getting heavy. Carry it into a new chat as a short brief and keep going.

Message Claude…

↓ re-read

skip the heavy thread

Nothing sent until you click

claude.ai

Help me plan a 3-day trip to Lisbon.

Here is a simple 3-day outline: Day 1 Alfama & the cathedral, Day 2 Belém & the river, Day 3 a local market and a sunset viewpoint.

Add food stops for each day — one lunch and one dinner near the places you listed.

↓ tokens

fewer wasted messages

Next moves

Up to four ways forward. Pick one.

When a reply finishes, CONTEXA offers up to four independent next moves — each a complete, ready-to-send message mined from your own side of the whole conversation. Click one and it lands in the box, unsent.

Four independent options — pick one, discard the rest
Each stands alone, complete and ready to send
Lands in the message box unsent — you edit and send

Measured, not claimed

The savings, in real numbers.

On claude.ai every message is processed with the whole conversation behind it — so on a long thread that cost is paid again on every send. Here is what that looks like, measured and modelled.

The same five messages, in a 689k-token session and in a new chat with its brief

Five messages, one after another	Five-hour limit spent
Continuing the 689k-token thread	9%
In the new conversation, with the brief	3%

One live session, one phone, same model on both sides, the usage page read before and after each set of five. Three times cheaper per message.

The brief against the thread it replaces, five sessions

Where it was pressed	Re-read per send	The brief
Cowork session	689k tokens	441 tokens
Cowork session	214k tokens	313 tokens
Cowork session	124k tokens	444 tokens
Chat	22k tokens	427 tokens
Chat	14k tokens	442 tokens

Five sessions, one day. The brief lands between 300 and 450 tokens whatever the thread was.

Modelled spend on one session, without and with CONTEXA

Session	Messages sent	Tokens read	Saved
10 asks, at 1.7 messages per ask today	17 → 11	146k → 60k	about 59%
20 asks, at 1.7 messages per ask today	34 → 22	596k → 247k	about 59%
40 asks, at 1.7 messages per ask today	68 → 44	2.4M → 1.0M	about 58%
40 asks, at 1.5 messages per ask today	60 → 44	1.9M → 1.0M	about 46%
40 asks, at 2 messages per ask today	80 → 44	3.3M → 1.0M	about 70%

A model, not a measurement, and it says so: an ask is one thing you want done, each exchange taken as ~150 tokens from you and ~900 from Claude. The percentage barely moves with length; the absolute saving is what grows. The full method, with every caveat, is on How it works and in the token-savings thesis.

How it works

Three steps. No new habits.

CONTEXA works right inside your Claude chat. You stay in flow — it just hands you the next move.

Claude replies

When a reply finishes, CONTEXA quietly appears above the message box.

You get a menu

Up to four good next moves show up as clickable chips — read for context.

Pick one

Your choice lands in the box, unsent. Edit it if you like, then send.

Features

Small tool. Big on respect.

Everything you need to keep a conversation moving — and nothing you don't.

Privacy-first by design

CONTEXA reads the on-screen conversation to shape suggestions. Nothing is sent anywhere until you actually click a move. Your choice, your call.

No tracking No hidden calls You decide

Up to four next moves

Independent options based on the chat. Pick one, discard the rest.

Free to use

No paywall for the core workflow. Just install and go.

No account needed

Nothing to sign up for. It works the moment it's installed.

Optional API key

Add your own personal key to remove usage limits.

Benefits

Why people keep it on.

CONTEXA makes each reply count — so your conversations go further on fewer tokens.

Spend less

Save tokens

Fewer wasted messages and fewer reruns on long threads.

Ask better

Better follow-ups

Clear, single-purpose prompts that keep your context intact.

Move faster

Stay in flow

No tab-switching. No prompt library. Just pick and go.

Keep your Claude sessions moving.

Install CONTEXA and get a menu of smart next moves — one click at a time.