Caveman
Token optimization stack that cuts AI agent costs and context size
Introduction
Caveman reduces the cost and context size of agent-based development through compression, caching, routing and spend tracking. It ships an MIT-licensed skill and a free local proxy that runs with your own provider keys. Telemetry is limited to token counts.
- Added on
- Nov 2, 2026
- Website
- caveman.so
- Pricing model
- Freemium
Product information
What is Caveman?
Caveman is an efficiency stack for agent-native development, made up of a skill, a local proxy and an Agent SDK, with cloud and enterprise services planned.
How to use Caveman?
Install the free Claude Code skill with the provided shell command, or install Caveman Code and Cavemem via npm. For the local proxy, start an agent through the wrapper, for example "caveman claude". An account is optional.
Core features of Caveman
- Recoverable compression of logs, JSON and code
- Token usage and savings visibility
- Caching, model routing and rollback
Use cases of Caveman
- 1Cutting token costs with Claude Code, Codex and Cursor
- 2Compressing large logs and tool outputs
- 3Monitoring AI spend by member, key and model
- 4Testing cheaper model routing
Frequently asked questions about Caveman
Can I use Caveman without an account?+
Yes. The local wrap runs on your machine with your own provider keys and needs no Caveman account.
What does the Caveman free plan include?+
A one-seat local wrap, the MIT-licensed skill and extension, locally computed inferred savings and optional cloud sync.
Does Caveman send my prompts to its servers?+
The local wrap states its telemetry contains token counts only and never prompts.
Which category is Caveman in?+
Caveman is listed under AI Code Editor.