Reduce Claude Code token usage

Same work.
Fewer tokens.

Most wasted tokens are not in your prompt. They are in the PDF you attached, the web page you pasted, the screenshot at full size and the session that has dragged its whole history along for hours.

10 available now 1 in development

AigoPilot cuts that waste where it starts, before the model reads anything, and shows you the result in numbers instead of promising a percentage. Under every file you see its tokens before and after. The front page of aigopilot.com, for one, went from 9,806 tokens of HTML to 2,911 tokens of text.

01Available now

PDF, Word, Excel and PowerPoint, turned into text

Attach a file and it is turned into Markdown on your own computer, with no internet connection and no outside service. Headings, lists and tables keep their shape: a table in a PDF comes out as a table, even when it carries on over two pages. Spreadsheets come out without the usual junk (no NaN, no Unnamed columns, no stray .0), and hidden sheets are named as left out.

Each file is turned in a separate process with limits on memory and time, so a huge or broken file cannot freeze AigoPilot.

02Available now

The tokens saved, under every file

Under each attached file one line says what its text costs compared with the file itself, for example “its text: 2,911 tokens instead of 9,806”. If something could not be read, that line says what is missing, in red. Click it to open the text and check it yourself.

03Available now

Web pages, reduced to their main text

Paste an address and AigoPilot fetches the page once and keeps only its main text: no menus, cookie notices, side columns, share buttons, comments or footer. Headings, lists, tables and code stay as they were, and links are kept whole. An address that points to a PDF or Word file is turned like any other file.

A page behind a “prove you are human” wall is refused with the reason, after two more tries, and never saved as if it were the content. You are told how to bring it in by hand instead.

04Available now

A file is turned only once

The text of a file is kept next to it. Attach the same file again and, if its bytes have not changed, the text from last time is used, with no new work and no new cost.

05Available now

Images shrunk before they reach the model

Every picture is resized to under 2000 pixels on its long side, however it was added, and the original is kept. If the model refuses an image, the message says which one and why.

06Available now

See what fills the context before you type

The context bar shows how much the system prompt, tools, MCP tools, memory files and skills take before your first word. An MCP server you never use, or a memory file that has grown out of hand, shows up right there, so you know what to switch off. More on the context bar.

07Available now

Hand over instead of dragging a full context

Every new turn of a long session sends its whole history again. One click writes a handover note and opens a fresh, linked session that starts from the note instead, and the button shows what the handover costs before you press it. For Claude Code, and since October 2026 for Codex too. How the handover works.

08Available now

A warning before an expired cache is read again

A session left idle longer than its prompt cache lasts will read its whole context again at full price. AigoPilot warns you before you send, with the number of tokens involved, so you can start a fresh session instead.

09Available now

Compact, hand over or switch when it is full

When a context fills up, a card offers three ways out: compact it, hand it over, or move to a model with a bigger window. You pick, instead of the session stalling.

10Available now

Find the project that eats your limit

Each project shows its share of this week's usage, and each session what it took from the five-hour and weekly windows. The token ledger counts every model call once, sub-agents included. Usage and limits.

11In development

Prompt caching, compaction and sub-agents, in one place

One place for the other ways to get more done within the same limit: keeping the prompt cache warm, compacting the context at the right time and handing parts of a job to sub-agents. No particular saving is promised.

QUESTIONS

Before you
decide.

How can I reduce Claude Code token usage?

Send the model less noise. In AigoPilot, files and web pages are turned into clean text before Claude Code reads them, pictures are shrunk, the context bar shows what fills the context, and a long session can hand over to a fresh one instead of carrying its whole history.

Why does a PDF use so many tokens in Claude Code?

A PDF read as it is carries its layout, and a scanned page is a picture. Turned into Markdown first, only the words and tables are left. AigoPilot shows the tokens before and after under each file.

How many tokens will AigoPilot save me?

It depends on what you send. A plain text file saves nothing. A web page or a PDF full of layout can lose most of its tokens, as the front page of aigopilot.com did, going from 9,806 to 2,911. AigoPilot shows the real number under each file instead of promising one.

Are my files sent anywhere to be converted?

No. PDF, Word, Excel, PowerPoint and saved web pages are turned on your own computer with no internet connection.

MORE FEATURES

What else is in the app

More done
before the limit.

Get AigoPilot Next: Working with agents