Drag or flick in any direction.
Analyze Past Usage Your chosen history file is not sent to TOP
TOP-1: Icarus, being tested

Your past AI tasks should help price your next one.

TOP uses your past Claude and ChatGPT work to test a low, likely and high cost estimate for the next task. Today, you can privately turn a downloaded copy of your account history into a report on this device. The next-task estimate is still being tested.

Analyze my past usage Meet TOP 1, 2 and 3
THE TOP-1 VISION, NOT LIVE YET
Task: update sign-in and add tests
lower estimatemost likelyhigher estimate
€0.18
LOW
€0.42
LIKELY
€1.10
HIGH
future idea A later control could stop a task at your chosen spending limit. It does not work today.
WORKS NOW: see past usage  ·  BEING TESTED: estimate the next task  ·  LATER: help choose the right AI
what makes TOP different

A normal usage screen is a receipt. TOP aims to give you the quote first.

AI companies publish prices for small pieces of text. That does not tell you the price of finishing a real task. The AI may need several attempts, use other software or continue the conversation. TOP is testing whether your previous work can make the next estimate more useful.

Think of a taxi with no meter. AI work can feel like getting in without knowing the fare. TOP aims to show a sensible range before the journey starts.
live today

See how you have used AI

Choose a downloaded copy of your Claude or ChatGPT history. The report is created on your device. You do not need to change code or connect a developer account.

the idea we are testing

Estimate the next task before it starts

TOP is not the only product trying to estimate AI costs. The difference we are testing is a personal low, likely and high amount for a complete task, shaped by what your past tasks actually used. We still need to prove that this makes the estimate more accurate.

one system, three layers

Know the cost. Choose the right AI. Improve the work.

TOP stands for Token Optimization Protocol, meaning a system for getting more useful work from your AI allowance. Tokens are the small pieces of text AI services count when measuring usage. TOP-1 is what we are building now. TOP-2 and TOP-3 are ideas for later. Icarus, Daedalus and Athena are temporary names.

TOP-1 · Icarus · being tested

Know before you fly.

Icarus is the warning.

In the myth, Icarus ignored the safe limit and flew too close to the sun. TOP-1 aims to show a low, likely and high cost before an AI task begins.

Like a weather forecast for an AI task. It shows a likely range, not a promise of the exact bill.
Works today: a report about past usage. Being tested: the next-task estimate. We have not yet shown the complete estimate working reliably in a public test.
TOP-2 · Daedalus · idea for later

Choose a smarter route.

Daedalus is the inventor.

Daedalus built the wings and understood the safe way to fly. TOP-2 is an idea for comparing AI versions and finding the right balance of cost and quality for a task.

Like sat-nav choosing the route, not merely showing the fuel price.
Not available yet. We have not proved that it saves money.
TOP-3 · Athena · idea for later

Learn to work smarter.

Athena is the strategist.

Athena represents wisdom and strategy. TOP-3 is planned as a personal coach that asks useful questions, learns how you work and helps turn a rough request into a better plan.

Like a coach who knows how you work, not the same set of instructions for everyone.
Not available yet. We have not proved that it improves anyone's work.
why it matters

The advertised AI price is not the price of finishing your task.

A price list for bricks cannot tell you what a whole house will cost. An AI price list has the same problem.

01

One task can hide many steps

A short request can lead to planning, extra software, repeated attempts and a long conversation. The advertised price does not reveal the final bill.

02

Your habits change the cost

Two people can ask for the same result and use AI very differently. A useful estimate should learn from how your work actually runs.

03

The lesson arrives too late

A bill is a rear-view mirror. By the time it shows an expensive task, the money and time have already been spent.

the TOP-1 goal

From one task description to a sensible range.

This is how Icarus is intended to work. The report available today does not estimate your next task yet.

01

Describe the task

Say what you want in everyday words, such as fixing a sign-in bug, researching a market or drafting a report.

02

Compare it with the past

TOP aims to find similar completed tasks and use your own history when enough useful history exists.

03

See the range and decide

A low, likely and high figure should help you run the task, change the approach or wait. It is an estimate, not a guarantee.

available now

Start with what already happened.

Download a copy of your Claude or ChatGPT history and open it here. Your report is created on your device.

01 · bring your history

Choose a copy of your history

Use a copy downloaded from Claude or ChatGPT. You do not need to change code or connect a developer account.

02 · keep it private

Your file stays with you

The browser reads the history file on your device. That file is not uploaded or sent to TOP.

03 · see the pattern

Understand how you use AI

See how much AI you used, when you used it and which kinds of work used the most. Any cost estimate is labelled so it is not mistaken for a recorded bill.

Analyze my past usage →
No new account · no code change · your chosen file is not sent to TOP
where it stands

What is real today.

One tool works now, one feature is being tested and two ideas are planned for later.

Works now
A report made from a downloaded copy of your Claude or ChatGPT history, without sending the file to TOP
Being tested
TOP-1 Icarus, an estimate shown before the task begins
Later ideas
TOP-2 Daedalus and TOP-3 Athena are not available yet

The early estimate is not accurate enough yet. In a test using one founder's Claude Code history, most records came from the same repeated task. Across 63 later sessions from other work, the typical estimate was about 3.3 times what the same usage would have cost at standard developer prices, where each amount of AI use is billed separately. That tells us what to fix. It does not prove that TOP works for everyone.

a simple example for TOP-2

Cheaper is not better if the work gets worse.

Daedalus is an idea for helping people choose which AI version to use for each task. The cheapest option is useless if it gives worse work. In this made-up example, the green line rises twice as much as the red line: one step of extra AI cost corresponds to two steps of useful work. It is not a customer result.

TOP-2 IDEA, EXAMPLE NUMBERS ONLY
what the AI costs useful work, rising 2× as much invented example after the cost
Illustrative AI cost and useful work An invented example in which every rise in useful work is exactly twice the corresponding rise in AI cost. This is not a customer result.
low

Keep most of the saving while the value of the work keeps rising.

useful work for every €1 or $1 spent, before
~0.85
invented example, useful work for every €1 or $1 spent
5.4
invented example change
6.3×

Illustrative, not measured. These are example figures, not a customer's results.

Why a cheaper task may not mean a smaller total billWhen each task costs less, teams may run more tasks. The price per task falls, but the total bill can still rise. This pattern is sometimes called the Jevons effect.
privacy

Your files stay on your device.

The analyzer reads your chosen history file in your browser. That file is not uploaded to TOP.

✓
Opened only on your device. Your browser creates the report without sending the file to TOP.
✓
You decide what to share. The analyzer does not send your chosen history file to TOP.
✓
Your conversations stay out of the shareable copy. The copy leaves out conversations, replies, code and internal ID numbers.
✓
Partner collection is not open yet. We will not accept partner files until we have clear rules for permission, storage and deletion.
early test partners

Help us test TOP with real work from businesses.

We plan to invite a small group of businesses to help improve the estimate. We are not accepting their files yet.

you share

If you choose to join later: a summary you have checked yourself. It would show totals and broad patterns, but not your conversations or code.

you get

A private review of your AI spend, early product access, and a voice in what we build next.

AH
SO

Built by Adam Hartley, co-founded with Sam O'Connell

Adam and Sam are building TOP to make AI spending easier to understand and control. Adam leads the current product build. Sam helps shape the business, challenge the claims and test whether the explanation makes sense to a new reader.

Adam on LinkedIn ↗ Sam on LinkedIn ↗ Sam on GitHub ↗ adam1hartley@gmail.com +353 85 268 7653

See what happened before. Help us estimate what comes next.

Analyze your past usage privately, or tell us about the AI tasks you wish came with a quote.

Analyze past usage → Talk to the founders