Transcription
Imagine this. You're using cloth. You're brainstorming. You're prompting. You're getting some good ideas. And then you get this message. 5 hour limit reached, resets at 9 p.m.
If this keeps happening to you, it's not the plan. It's how you actually use it. I'm going to give you five different fixes. Rank from easiest to more advanced ones. And the first one actually takes zero setup and you'll probably solve most of your issues right away. The last one is oneline install that cuts your token use by 65%. So let's get it started.
Tip one, stop living in one long chat. Every time you send a message, Claude rereads entire conversation. The full history, every message before, every file you uploaded, every reply. So a 40 message chat cost way more per message than a fresh one. Basically, you're not just paying for your new question, you're paying to reprocess everything above it. And the fix is really simple. When you switch to new task, start a new chat. Finish writing an email, new chat. Moving to a different project, new chat. This one habit alone will probably solve most of your problems. And most people never do it because staying in one chat feels easier, but it's secretly burning all your limits and your tokens.
Tip two, pick the right model. Claude has three models. OPUS is the most powerful one. Sonet is the balanced one, and Haiku is the fastest. and cheapest. Most people default to Opus for everything. That's why they hit limits real fast. OPUS is really expensive to run. And here's the rule I use. If you wouldn't hire a senior consultant for the task, don't use Opus. For example, writing an email set. Coming with ideas, set simple question. Hi, Cool. Deep thinking or complex code. That's when you use Opus. You can switch models in the drop down at the top of the chat. Use Sonet for 90% of your work and you'll feel the difference in your limits immediately.
Tip three, use projects for anything you repeat. If you find yourself posting the same context into every chat, you're doing it wrong. Your brand guidelines, your writing styles, your company info, your instructions for how CL should respond. That's what projects are for. You upload your context files once. You write your instructions once. Then every chat you start inside that project already knows everything. You don't repaste anything. This saves tokens two ways. You're not sending the same paragraph every chat and your chat stays shorter because Claude has the context built in. I run my YouTube channel inside the project. All my brand info is there. Every script I write starts from the context. I never copypaste the same information twice. So if you only use claudon free plan projects are locked which is a bummer but if you're using pro or higher you can use them right now and they're the biggest upgrade most people ignore.
Tip four plan mode in cloth code. Now we're moving into cloth code. This is the tool a lot of people are burning their limits with. Here's the problem. You give clot code a task. It decides on its own how to spin it in five different sub agents, runs tools, reads files, all before it even starts answering to you, and you just watch your tokens disappear in seconds without your input. The fix is plan mode. Press shift tab before you send your prompt. You can also find plan mode next to the prompt window and basically type in what do you want to build and plan together with claude before executing anything and spending your tokens. This will give you way better output with way less token burn. So plan mode is the single biggest fix for clot code users.
Tip five, caveman mode. This one's actually wild. I found out about it on YouTube from other YouTubers. So this is basically a free skill on GitHub called Caveman made by developer named Julius Bruce. I hope I'm saying it right. You can also find it in the description. Basically, it's a oneline install and what it does is really simple. It makes Claude respond in caveman speak almost like me. Short sentences, no filler modes, no I'd be happy to help you, no hedging, no articles like a and or the and same technical accuracy, but up to 75% fewer output tokens in response. I will give you one example right now. So, normal clot says something like this. The reason your React component is rerendering is likely because you're creating a new object reference on each render cycle. I'd recommend using use memo to memorize the object. Kayman Claude says this new object ref each render inline object prop equals new ref equals rder and use memo. Same answer, way shorter. And the benchmarks on GitHub show average of 65% token savings across real coding tasks which is really impressive. Yes, it sounds like a joke. It's a little bit funny but the benchmarks they are real. So one honest note about the caveman caveman cuts output tokens not thinking tokens. So it will not magically fix your usage limits on its own but if you combine it with other tips mentioned in this amazing video it will add up fast. And if you code with clot every day, the time saved from shorter responses is huge and actually I think makes way more sense. Again, you can find it in description.
And now I have a bonus tip for people who are still watching. Also, don't forget to subscribe. If you use plot code, type /clear between tasks. Don't leave one session running all day across different projects. Clear the context when you switch. It will reset your token use. And if this video helped you, don't forget to subscribe. And please in the comments tell me what else do you want to see on this channel.