Caveman

Meet Caveman

Caveman is an open-source toolkit for reducing the amount of text AI agents read and produce. It compresses eligible tool results such as logs, JSON, diffs, and documents before they enter the model's context. Supported integrations preserve access to the original material when an agent needs exact details.

Developers can use it with coding agents or integrate its TypeScript and Python components into applications they build. Model-routing features select models and effort levels for supported workflows, with an account required for adaptive selection. Caveman Cloud adds usage visibility and evaluates potential optimizations against observed traffic. Savings depend on the task, model, and integration; teams can compare cost and output quality before deciding which changes to keep.