SpendwaiseRecorded live

Chat

ShopBot · 6 tools

Ask an agent a real question.

ShopBot works on a sample online-shop codebase. Its tools return big, noisy output: test runs, logs, JSON, web pages. Spendwaise trims each one before the model reads it.

Behind the scenes

ModelToolSpendwaise

Every step of the answer shows up here.

  1. The model asks for a toolIt decides what to read, and every call is priced.
  2. The tool returns its raw outputOften thousands of tokens of logs, JSON or HTML.
  3. Spendwaise trims it before the model reads itYou see what was kept, what was cut and why.

Savings

0 model calls
Saved on this conversation
–
With Spendwaise
$0.00000
Without
$0.00000
Tool output the model read

Tokens each tool returned → what the model actually read.

0→0tokens

Ask a question to see each tool here.

Input tokens sent
0 vs 0
Read from prompt cache
0% vs 0%
Where the money goes
with$0.00000
without$0.00000
system prompt + toolsconversationtool outputs, trimmedtool outputs, rawmodel output

Spendwaise only shrinks tool outputs. The rest costs the same either way, so the more an agent reads, the more it saves.