Caveman: FREE Tool Cuts Tokens by 69%!

So today we're going to be looking at a new way to cut 65% of tokens with Claude code. So if you're using Fable 5, this is an easy way to cut the amount of tokens you use, but still get the most out of it so you can use it even more. I mean, imagine if you can code 65% more with Fable 5, how much more would you get done? 65% more.

So that's why we are using this system to reduce the amount of tokens that we use when we are running AI agents. And you can also use this with any other agent, you know, it's not just for Claude code. You can also run it with Codex, Gemini, Cursor, Windsurf, Klein, Copiler, and it's free and ready to go, as you can see right here. You also might say, how does this work?

Well, I'll run through that in a second. And if you're wondering, okay, how to install it, you can just copy and paste this command and give the whole setup to Claude code so that it can run with Fable 5. And the great thing about this is you can install it for one agent or even for 30 different agents, which is pretty insane. And the way this works is basically, you talk to your agent like a caveman.

And when we actually tested it on Fable 5, it used 69% fewer output tokens. I'll show you the example test in a minute. It was 37% cheaper, and every answer was still pretty good. So here's the problem with, for example, you know, something like Fable 5.

It uses up a lot of tokens. When it stops being free on the subscription and you have to pay per API, it's going to get quite expensive if you're not careful. And so caveman is a free little skill that fixes the talking, not the thinking. And you install it once, and then your agent basically answers like a smart caveman, short, blunt, and right.

And then the way this works is basically, you know, the idea is that when an AI answers you, you are using up tokens for every word it says. And those are output tokens. And on the smartest models, they're quite expensive. They use up a lot of tokens.

With caveman, it's a set of instructions your agent reads once at the start of every chat. And the instructions say, for example, drop the filler words. No, like, for example, sure, I'd be happy to help. No, basically and actually.

No, three paragraph warmups. You say the thing, you give the fix, and it stops talking. And there's basically three rules inside caveman to make it use less tokens. So number one is code is sacred, right?

So code blocks, commands, file paths, error messages are never touched. Then also, it knows when to speak up. So obviously, sometimes it will need to speak more in situations related to security. And so this way, it knows, okay, during those situations, we're going to speak more.

And also, it can speak multiple different languages. If you, for example, write in Portuguese, it will reply like a caveman in Portuguese. It compresses the style, but not the actual meaning. And so the way this works is it thinks exactly the same.

It's still thinking like it would normally, but the difference is the filter afterwards. So all the filler words that come back to you are reduced. And so you use less output tokens when it's replying back to you, when it's coming back to you inside the chat. Also, you might say, okay, you know, aren't shorter answers when it's replying to you, like making it worse?

Well, basically, apparently, sometimes this can be the opposite. So March 2026, research paper tested 31 models and found forcing brief answers improved accuracy by up to 26 points on some benchmarks. And in my own five-prompt test, every caveman answer contained the same fixes along the version, right? So it was still working just as well.

It's just like it replies with less words, and it's a bit shorter and more straight to the point. And if you're wondering, okay, what it actually is and what happens in every message, basically, it's just a text file full of rules. So that's the whole secret. It's not like an app or a new model or a middleman or anything like that.

It's just like a written instruction, a skill that tells your agent how to speak, nothing more. Now, the installer basically copies that file into your agent. So you run the one-line install. It looks at your machine, finds every agent you have.

And then with Claude Code, Codex, Gemini, 30 others, it drops the same rules file into each one's skill or plugin folder in the format that agent expects. And then also, your agent reads the rules at the start of every chat. So AI agents always read their instruction files before answering you. That's how skills work.

So for message one, the caveman rules are sitting in the model's context right next to your question. Now, also, the rules tell it what to stop. So the actual file says, almost word for word, drop articles, drop the filler, drop pleasantries. So it's not going to say, like, happy to help, sir, thank you, et cetera.

And also, it answers in the pattern, which is thing, action, reason, next step. Also, the rules tell it what to protect, like we were talking about before. And the model writes, sure, and it never edits afterwards. So it's not going to, like, trim the answer.

It just responds in a short way straight off the bat. So here's an example. If you ask, like, OK, why is my React component re-rendering? The agent bundles your question in the caveman rules file.

The model reads the rules, so it thinks exactly the same way. It just chooses fewer words as it writes, and you get a short reply. So instead of getting, for example, like 1,349 tokens used on the answer for the output of this, it would use, like, 324 tokens in our test. So we actually ran this with five different tests using Fable 5.

And it's pretty interesting in terms of the amount of tokens saved. On every single one, we saved a lot of tokens. We actually ran five real questions from the GitHub directly to Fable 5 using the caveman skill. It was pretty interesting.

So, you know, the normal out, for example, for, like, a React re-render bug, that would be normally, like, 1,349 tokens. But with this, it was 324. And you can see here, like, on the five tests, we saved a lot of tokens on every single one. The average was about 69% saved, which is pretty insane.

So 69% fewer output tokens, which would save about 37% of the total, if that makes sense, because this is just for output. It's not for input. And you can use it and install it with one single command, which is this one right here. You could also use it inside your agent operating system.

So, for example, if we were using Claude, and this is actually something that's really, really good, is, like, because you've got so many agents plugged into the agent operating system, if you're using multiple agents like Hermes, like OpenClaude, like Claude, like Antigravity, you can install Caveman across all of them. And then it's reducing the amount of tokens across all of them. And so, like, it has a massive compound effect because you're saving 69%, not just across one CLI, but across every single agent and CLI that you use, which is pretty amazing in itself. And we checked the answers were right.

So all of the answers were much shorter, as you can see, but also they were all correct. So the great thing is you actually get the right answer back. Now, this really matters on Fable 5 because, obviously, number one, you've got limited tokens on the subscription plan. Number two, when the subscription plan on the 7th of July drops, well, at that point, you won't be able to use Fable 5 on the subscription.

You have to use the API. If you're using the API, you want to reduce the amount that you use the API. Otherwise, you can get expensive. So this way, you can just shave off a lot of tokens and save a lot.

So let me show you, like, the old way versus the new way of this. Like, the old way would be, for example, every reply opens with, like, sure, I'd be happy to help with that or something like that, right? And then you've got three paragraphs to set up before the actual fix. You scroll and skim to find the one line that actually matters.

We've all seen that inside, for example, Claude as well. Like, you can see quite often it will give you a long answer like this, but you can reduce it massively by using Caveman. And then also, the great thing about this as well is, like, quite often in memory files, when you're using Claude code and that sort of thing, the replies are trained to be quite buggy or quite full of too many rules, if that makes sense. Whereas with this system, it uses 69% fewer tokens on my own tests with Fable 5.

You read the reply in one glance, which is faster than skimming because it's much shorter. Everything stays intact. So all the code, the commands, everything else you have. And then the memory file is compressed once, which means that once it's compressed, the savings repeat.

And it's one free install, right? It's a free and open source project. And you might also say, like, I'm not technical, can I actually install this? It's literally one command.

So if we scroll up, all you need to do is use that. And if you don't know how to do that, you can just give the command to Git, you can take the GitHub and then give it to Claude code and that can set it up for you as well. Now, you can also switch this off at any time too. So you can do forward slash caveman to bring it back at any point.

And you can also pick how it replies. So you can switch between light, full and ultra. So if you don't want to use this all the time, no problem. If you want to use like some of the bonus commands, like caveman commit, for example, or for example, caveman review or caveman compress, these are all different options.

You can reduce the token count even more as well. And you can see how the responses vary. So if we look at the system, if you're using like the normal, it would reply like this. If you're using light, it would reply like this.

With full, like so. And with ultra, even shorter. So you can make the response shorter depending on which setting you have. If you're wondering how to change the setting, you just change it like so.

So that's basically it. That is how I saved 69% of my tokens on Fable 5 using this system. Pretty powerful. If you want the full token saving stack, caveman shrinks the replies.

The agent operating system is designed for that as well. So it can work with all of this. The other cool thing is that we have free local models inside the agent operating system. So you can use, for example, the local system here.

And we also have OmniRoot plugged in as well, which means that you can code for free using a system that delegates everything to free APIs as well. On top of that, we have loads of cool features inside the agent OS, like Hermes with Hermes Oracle, Hermes Astros, a voice agent called Apollo. We have Claude that can code directly. And we have a memory system plugged in as well.

So this is a powerful system to make the most of caveman so that you can use even more tools, but reduce the amount of tokens. We have loads of token efficiency playbooks as well inside the AI Profitable Boardroom. So if you want to get that, feel free to get it inside the AI Profitable Boardroom. Link in the comments description or go to the AI Profitable Boardroom.com.

Inside the community, you can ask questions and get help and support in real time. And I personally answer every single question inside there with a video tutorial. Inside the classroom, you can get access to all of my best trainings, including the agent OS system that we update daily, get a video tutorial, full guide on it, and the zip file to install it. We also add new daily guides based on what's actually useful that has just dropped.

And I test all of this stuff personally, as you've seen today. So for example, today we tested out five different experiments. We do that with every single tutorial inside the AI Profitable Boardroom. You get the video tutorial, the full step-by-step guide, and everything you need to win with this stuff.

Inside the calendar, you can jump a weekly coaching call, share your screen, ask questions in real time. Inside the map, you can meet people in your local area who are building with AI agents like you, and you can connect with me personally. So feel free to get that link in the comments description or go to the AI Profitable Boardroom.com. Thanks for watching.