
Caveman
Open-source skill, CLI and SDK that cut tokens AI agents read and write, plus a hosted Cloud for agent spend.
Caveman is a free, open-source skill, CLI and SDK that make AI agents such as Claude Code and Codex say less and read less, plus Caveman Cloud, a hosted service for tracking and cutting agent spend. The local tools need no Caveman account; Cloud is free for one person, and Team costs $300 a month with $300 of usage included.
About Caveman
Caveman started as a joke that made agents talk like cavemen. The skill, which your agent loads, takes the filler out of answers and leaves code, commands and errors byte-for-byte exact; the CLI shrinks logs, test output, JSON and diffs before the model reads them, and the original stays on disk if the model asks for it. Caveman says it wraps Claude Code, Codex and Gemini CLI, that shorter replies also work in agents such as Cursor, and that the skill covers 30+ agents.
For agents you build, the SDK sits where tool results become model input, as middleware for LangChain, the Vercel AI SDK and the OpenAI and Anthropic SDKs in TypeScript or Python. A local runtime runs beside your app, eligible logs, JSON, diffs and documents shrink before the model call, and in supported agent loops the model can ask for the exact original. Your framework, provider, keys and saved history stay the same.
Caveman Cloud, aimed at the company paying for the agents, runs hosted at app.caveman.so or in your VPC on Enterprise; the docs describe it as a managed LLM gateway with telemetry, evaluation and improvement tooling. It splits one bill by agent, model and day, records every request with its cost, turns your own requests into tests, and tries each suggested fix on old requests first; Autopilot tries fixes on a little traffic and undoes bad ones, with your approval.
Cloud Free gives one person 100,000 events, 600 agent minutes, 25,000 routing decisions and $10 of Caveman AI a month with no card, and adding a card lets usage go past those allowances at standard rates. Team covers your whole team with no seats, starts with a 7-day free trial and adds a semantic cache; on Team, SSO comes with the Scale package at $750 a month. Enterprise is an annual contract priced on the AI spend Caveman manages, and can run in your VPC on AWS or GCP.
Sources: caveman.so, /pricing, docs.caveman.so, /products, /products/caveman, /products/cavemem and /products/caveman-code, read 2026-10-26
Caveman pros and cons
Pros
- The local CLI, shorter replies and context compression work without a Caveman account, and the agent and SDK compress on your machine, keeping the original bytes there.
- Token markup is 0%: model calls run on your own provider keys, and your provider bills you directly.
- In a study Caveman cites, JetBrains' Denis Shiryaev measured 8.5% fewer output tokens over 82 paired coding tasks with no detectable quality difference, though total cost varied.
- Cloud has no seats on any plan, so Team covers your whole team for one fee.
Cons
- Savings vary: Caveman says its input and output figures come from separate studies that do not add together, and a flat monthly subscription may cost the same.
- On the Free Cloud plan, Caveman may use your prompt text and recorded requests to improve its models; moving to Team turns that off.
- Support for shorter replies and for context compression differs by agent, so Caveman says to check the setup guide for yours before installing.
- Team's included usage does not roll over, and past it you pay standard rates up to billing limits you set.
Tagged
- Web-based
Caveman alternatives
The closest alternatives to Caveman.
Caveman pricing
Read off the vendor’s own pricing page on 2026-11-02.
| Plan | Price | Best for |
|---|---|---|
| Free | $0 / month | One developer trying everything Caveman does, or a solo founder evaluating the service. |
| Team | $300 / month | Teams that want agent costs cut, with proof, plus project and content-sharing controls. |
| Enterprise | Custom / year | Organizations that need an annual contract and can deploy Cloud in their own cloud. |
Full Caveman pricing breakdown — limits, hidden costs, change log →
Related reading
Featured on Vibedonalds
Own Caveman? Add this badge to your site to show you’re listed — and link back to your profile here.
<a href="https://vibedonalds.com/tools/caveman" target="_blank" rel="noopener nofollow">
<img src="https://vibedonalds.com/badge/featured-on-vibedonalds.svg" alt="Caveman — Featured on Vibedonalds" width="240" height="60" loading="lazy" />
</a>Frequently asked questions
- What is Caveman?
- Caveman is an open-source skill, CLI and SDK that cut tokens AI agents read and write, plus a hosted Cloud for agent spend.
- How much does Caveman cost?
- Caveman pricing, read from the vendor's page on 2026-11-02: Free — $0 / month; Team — $300 / month; Enterprise — Custom / year.
- What platforms does Caveman support?
- Caveman's own site lists it for CLI.
- What are the alternatives to Caveman?
- The closest alternatives to Caveman are Edgee Claude Code Compressor V2, Paritok, Navige.
- What are the downsides of Caveman?
- Savings vary: Caveman says its input and output figures come from separate studies that do not add together, and a flat monthly subscription may cost the same. On the Free Cloud plan, Caveman may use your prompt text and recorded requests to improve its models; moving to Team turns that off. Support for shorter replies and for context compression differs by agent, so Caveman says to check the setup guide for yours before installing. Team's included usage does not roll over, and past it you pay standard rates up to billing limits you set.
- What is the Caveman skill for Claude Code?
- It is a skill your agent loads: it takes the filler out of answers and leaves code, commands and errors byte-for-byte exact. You add it with npx skills add JuliusBrussee/caveman, and Caveman says the same skill works in Claude Code, Codex, Gemini, Cursor and 30+ more agents.
- Is Caveman free?
- Yes: the agent and the SDK are free and open source and need no Caveman account, though you still pay your model provider on your own keys. Caveman Cloud's Free plan is $0 for one person with no card; Team costs $300 a month and includes $300 of usage.
- Will Caveman lower my Claude bill?
- It depends on how you pay. Caveman says fewer billable tokens can reduce metered API spend, but cached input, retries and recovery calls affect the result, and a flat monthly subscription may cost the same. It suggests comparing total cost per successful task.
- Is Caveman on GitHub, and what is the licence?
- Yes: engineers have filed issues and opened PRs on its public repo. The pricing page says the skill, CLI, SDKs, middleware, extension and Browse are Apache-2.0, while Caveman Code and the standalone cavemem package are MIT.
- What happened to Caveman Code?
- Caveman Code, a terminal coding agent, has been frozen since August 2026: still installable and working, but no longer in active development. Cavemem and Cavekit are in the same state, and the Cavemem page says its compressed-memory core now ships inside Caveman.