caveman
114kCuts token spend by roughly a third. It works, but the core ships under a non-free licence — read the terms before you build on it.
We checked the published benchmark against our own run: token spend fell by roughly a third, and the figures match.
In numbers
less token spend
What it is
caveman trims what the agent reads out of the project before it ever reaches the model. As a result token spend drops by roughly a third.
The idea is that the input is where saving pays off: the less noise gets into the context, the cheaper every agent step. It suits anyone who pays for tokens or keeps hitting limits and wants to stretch them without switching models.
How it works
context
Trims what the agent reads out of the project before it goes to the model.
Know before installing
The core ships under a non-free licence. On a commercial project that is a question for a lawyer, not a developer: read the terms before you build on it.
Read the breakdown
We went through this tool in detail — what it does, where the catches are, and who it suits.
Comments