Savings guide · Version 0.3.0.32
How Token Saver saves money on AI tokens
By Genoskwa Enterprises Inc. · Published (America/Toronto)
Many AI assistants bill or limit usage by tokens, and tokens come from what they read. A long PDF, a full spreadsheet, or a big recording can cost far more tokens than the few facts you actually need. Token Saver prepares a lighter version on your computer first, so your assistant reads less.
This guide explains where the savings come from in Token Saver 0.3.0.32 and how the in-app dollars-saved figure is estimated. If you are new to the app, start with what Token Saver does.
Fewer tokens in, lower cost
Every character your assistant reads becomes input tokens. Token Saver shrinks that input before it leaves your computer:
- Recordings become transcripts. Audio and video are transcribed locally, and Send attaches the transcript instead of the original recording.
- Documents become extracts. PDFs, docs, and other supported files can be read as an outline, selected pages, or just the text.
- Spreadsheets send only what changed. Change tracking lets your assistant ask for the rows that changed instead of re-reading the whole sheet.
- Big visuals get lighter copies. When you need the picture, you can make a smaller video or image copy. Audio files such as WAV stay transcript-only.
- Approved answers can be reused. You can share individual memory facts and answers you already approved, so they don’t have to be worked out again.
How the dollars-saved estimate works
The savings counter in the app is a rough estimate, not measured billing. It works like this:
- Tokens avoided are the original file’s characters divided by 4, minus the extract or transcript characters divided by 4. Four characters per token is a rule of thumb, not any assistant’s actual tokenizer.
- Estimated dollars saved are those tokens times $2 per million tokens, the published public API input rate, rounded to the nearest cent.
Illustrative arithmetic, not a customer result: if a document has 400,000 characters and the extract you send has 20,000, the estimate is 100,000 minus 5,000, or 95,000 tokens avoided. At $2 per million tokens, that shows as about $0.19.
What the estimate is not
- It isn’t a SuperGrok refund or a change to any subscription bill.
- It doesn’t use output-token pricing.
- It isn’t a measured savings percentage. We don’t publish percentage claims.
Why saving tokens also helps on a subscription
Even when you don’t pay per token, assistants have usage limits. Sending a transcript or an extract instead of the whole file leaves more room for the work you actually want done.
Privacy comes with the savings
The same steps that cut tokens also keep originals local. Transcription and smaller copies are made on your computer, and original files and recordings don’t leave on their own. Only what you approve or send, such as a transcript, an extract, or a change list, can leave. Hosted extracts expire after at most seven days. Read the privacy details.
Try it
Token Saver 0.3.0.32 is available for Windows 10/11 (64-bit) and Mac with Apple silicon. It works with Grok Bot through a hosted connector, with no API key needed. Download Token Saver, or see Early Access pricing.