---
title: "Caveman is live on GreenPT"
description: "Until today there was one way to get caveman: install it yourself. The skill"
canonical: https://caveman.so/news/caveman-live-on-greenpt
last-updated: 2026-08-03
---

# Caveman is live on GreenPT

Until today there was one way to get caveman: install it yourself. The skill
file went into your agent, the agent started answering in caveman, and your
output tokens dropped. That loop has run roughly ninety-five thousand GitHub
stars deep, and every one of those installs was somebody doing it by hand.

GreenPT just removed the by-hand part. They've built caveman into their
platform at the API level: you call their API, turn compression on, and the
responses come back already speaking caveman. There's no skill file to install
and no prompt surgery; when the skill improves, there's nothing on your side to
update. It's the first platform to ship caveman as a built-in rather than an
add-on, and it applies to the traffic you send through their API.

## The numbers stay the numbers

Same benchmark we quote everywhere else on this site. Across ten prompts the
skill cut output tokens by 65% on average, with a spread from 22% to 87%
depending on how chatty the task was to begin with. Output only; input and
reasoning tokens don't change. Code, commands, and errors come through
byte-for-byte, because a compressed answer you can't paste into a terminal is
worthless.

And to be precise about what we have not measured: GreenPT's production
traffic. We won't quote a number for their platform until someone measures one
there.

## Why GreenPT, specifically

They sell green AI. Every token a model never generates is inference compute
that never runs, so on a platform whose pitch is cleaner compute, compression
stops being a cost trick and starts being the product doing its job. That's
also why the integration went deep instead of sitting on top: it lives at the
API level, where it covers whatever you route through them.

<Pull>Every token a model never generates is compute that never runs.</Pull>

<NewsCta
  kicker="Try it"
  body="GreenPT gives you a free API trial to test with. Turn compression on, send your own workload through, and read your own before and after."
  href="https://account.greenpt.ai/onboarding?signup=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWJzY3JpcHRpb25JZCI6ImFwaSIsIm92ZXJ3cml0ZVRyaWFsRGF5cyI6LTEsInRyaWFsT25seSI6dHJ1ZSwicmVkaXJlY3RUb0xvZ2luIjpmYWxzZSwic291cmNlIjoianVsaXVzLWNhdmVtYW4iLCJhcGlDcmVkaXRzIjoyfQ.upSp53KYZJRSxz_iWHcMNcZmlmMjffxB7J5JUaidQA4"
  label="Start on GreenPT"
  disclosure="referral link · if you become a paying customer, GreenPT pays us a share"
/>

We handle money claims the way we handle savings claims, so the disclosure sits
right on the button: that's a referral link, and it earns us a cut. If you'd
rather read before you sign anything, the integration docs are at
[docs.greenpt.ai/token-compression/caveman](https://docs.greenpt.ai/token-compression/caveman).

## If you run a platform

We built this integration once already; the second one is faster. Caveman drops
into a platform as a measured optimization layer without changing where your
customers work, and every published result stays honest: an integration is not
verified savings, and live numbers publish only after quality gates clear.
The [partners section](/#partners) is the door, or write to
[contact@caveman.so](mailto:contact@caveman.so).