---
title: "Caveman products"
description: "Caveman Code is our coding agent. The open-source Caveman CLI makes the agent you already run cheaper, and Caveman Cloud is where they run."
canonical: https://caveman.so/products
last-updated: 2026-11-12
---

# Caveman products

Caveman Code is our coding agent. The open-source Caveman CLI makes the agent
you already run cheaper, and Caveman Cloud is where they run.

| Product | What it is | Get it |
| --- | --- | --- |
| [Caveman Code](https://caveman.so/code) | The coding agent. Claude Agent, Codex, Cursor and Caveman, our own agent, in one app for Mac and the browser. Free on the subscriptions and keys you already have. | [Download for Mac](https://caveman.so/download) or [sign in](https://app.caveman.so/signup) |
| [Caveman Cloud](https://caveman.so/cloud) | Where Caveman Code runs, and where a team sees what every agent costs: spend traced to the work, evals on your own tasks, budgets and controls. Hosted or in your VPC. | [Sign up](https://app.caveman.so/signup), free for one person |
| [Caveman CLI](https://caveman.so/caveman) | Open source (Apache-2.0). Wraps the coding agent you already run, such as Claude Code, Codex or OpenCode, and makes it cheaper where it runs. | `npm i -g @caveman-ai/cli && caveman setup --install` |

## What the open-source tools change

| Tool | What it does | Measured |
| --- | --- | --- |
| [Caveman Skill](https://caveman.so/products/caveman) | Shorter replies with the same intent, in every agent that loads skills. | 65% fewer output tokens across 10 prompts, output only |
| [Caveman Proxy](https://caveman.so/products/caveman-proxy) | Logs, JSON, code and diffs shrink before the model reads them. The model can ask for the original. | 33.2% fewer input tokens: 6 Claude Code tasks, 3 paired runs each, 18 of 18 right in both |
| [Caveman Browse](https://caveman.so/products/caveman-browse) | A web page reaches the agent as the content it needs, not a full accessibility snapshot. | Up to 129.8 times smaller, best case: a 200-row table against a Playwright snapshot |
| [Caveman Middleware](https://caveman.so/products/caveman-middleware) | Local tool-result compression for TypeScript and Python agents. Version 1.0. | Savings inside apps are not measured yet |
| [Router](https://caveman.so/router) | Picks the model and effort for each turn in Claude Code. | 0.46 times the cost of always running Opus 5.5 on 85 paired tasks, same solve rate, interim |
| [CaveGemma](https://caveman.so/products/cavegemma) | Caveman compression baked into Gemma's weights. | Research |

Numbers measured on your machine are labelled inferred. Only Cloud reports
verified savings.

## Earlier tools

[Cavemem](https://caveman.so/products/cavemem) and
[Cavekit](https://caveman.so/products/cavekit) are still on GitHub without
active development.

[Developers](https://caveman.so/developers) ·
[Enterprise](https://caveman.so/enterprise) ·
[Pricing](https://caveman.so/pricing)
