Caveman
Open-source layer in front of coding agents that cuts tokens by shortening replies, compressing context and picking the model per turn.
Overview
Caveman sits in the request path of coding agents such as Claude Code, Codex, OpenCode and more than 30 others, and reduces token use by shortening replies, compressing context and choosing the model per turn.
The skill, local proxy and the trim-1 context-trimming model are open source under Apache-2.0, and the local CLI and proxy are free. Caveman Cloud is a hosted paid layer that prices every agent run and tests cheaper models, with a free plan for one person, a Team plan at $300 a month and custom Enterprise plans in your VPC.
Features
- Works with Claude Code, Codex, OpenCode and 30+ other agents.
- Shortens replies and compresses context.
- Picks the model for each turn.
- Includes trim-1, a small context-trimming model.
- Caveman Cloud prices every agent run and tests cheaper models.
Details
- Website
- caveman.so
- Company
- Caveman
- Category
- Agent infrastructure
- Source
- Open source
- Repository
- github.com/JuliusBrussee/caveman · 113.5K stars
- License
- Apache-2.0
- Deployment
- Local, Cloud
- Commercial offering
- Yes, paid offering available
- Topics
- Listed
- October 2026
- Last verified
Similar software factories
Encore
Open sourceOpen-source backend framework and CLI with a managed cloud that provisions infrastructure for agent-built code.TrueFoundry AI Gateway
Enterprise gateway that controls, logs and meters every AI call to models, agents and MCP tools.Agentic Factory
Open sourceInfrastructure, container images and reusable GitHub Actions for autonomous AI coding agents.Aiven
Managed open-source data platform that also hosts AI agents through Aiven Managed Agents and Aiven Runtime.
