Eight free plugins and habits that cut Claude Code's token bill, kill chat-planning chaos, and replace typing with talking, none of which depend on picking a smarter model.
Posted
yesterday
Duration
Format
Listicle
educational
Views
12K
305 likes
57 · 43
Big Idea
The argument in one line.
Working faster and cheaper with Claude Code comes less from picking a smarter model and more from a stack of free plugins and habits that cut wasted tokens, visualize plans before building, and reduce time spent managing chats and typing requests.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already use Claude Code regularly and want to cut your token spend without switching models.
You plan complex builds inside a chat window and keep losing track of decisions made several messages ago.
You're curious how Anthropic itself structures AI skills across a business and want a pattern to adapt.
SKIP IF…
You're brand new to Claude Code and haven't yet hit a token limit or a planning-in-chat problem, this is an optimization list, not a starting point.
You're looking for a comparison of which AI model is smartest, the video explicitly says that isn't the point.
TL;DR
The full version, fast.
A creator runs through eight free tools that make Claude Code cheaper and faster to work with, without touching which model you use. Lavish turns a build plan into a commentable visual document instead of a chat thread. RTK and Caveman/Ponytail attack token spend from both ends: RTK filters command output before Claude reads it, Caveman and Ponytail trim Claude's own replies and code. The new Claude Code Projects feature runs one conversation as parallel cloud sessions instead of juggling separate chats. BuildPartner.ai pulls live social content into Claude, Obsidian is a free local markdown viewer, Anthropic's Small Business plugin models how to structure skills for a company, and Hex replaces typing with system-wide dictation.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Turns a complex build plan into a rendered, commentable document instead of a scrollable chat thread, so a revision is a comment and a 'send to agent' click.
01:25 – 04:32
02 · RTK — stop paying for noise
A filter that strips irrelevant command output before it reaches Claude, attacking token spend from the input side.
04:32 – 06:41
03 · Caveman + Ponytail — write less
Two plugins that trim Claude's own output: Ponytail constrains code to the simplest working version, Caveman strips filler language from responses.
06:41 – 08:39
04 · Claude Code Projects — not what you think
A newly announced feature, reusing the old 'Projects' name, that runs one conversation as parallel cloud sessions instead of juggling separate chats.
08:39 – 10:02
05 · BuildPartner.ai — finding the tools
A connected research tool that pulls live content from Twitter, Instagram, and YouTube into a Claude terminal session, plus a weekly recap habit to avoid trend-chasing stress.
10:02 – 10:50
06 · Obsidian — reading what Claude writes
A free local markdown viewer for the files Claude generates, kept separate from a team tool like Notion.
10:50 – 12:17
07 · Anthropic's Small Business plugin
Anthropic's own set of business skills, like a sales pipeline tracker and inbox manager, used here as a reference for structuring skills across an organization.
12:17 – 14:50
08 · Hex — talking instead of typing
An open-source, system-wide dictation tool triggered by a hotkey, framed around two non-obvious benefits: it surfaces new ideas mid-thought and doubles as a speaking coach.
Atomic Insights
Lines worth screenshotting.
Claude defaults to dumping an entire command's raw output into context, so a single check can cost 15,000 tokens when only about 1,800 of them actually matter.
RTK (Rust Token Killer) strips irrelevant command output before it reaches Claude, with the project claiming a 60-90% reduction on what commands send back, not total session token usage.
A rendered, commentable plan beats planning inside a chat thread because you can leave a note like 'switch tool order' instead of re-explaining the whole request from scratch.
Ponytail's documentation claims roughly 54% less code (up to 94%), about 20% cheaper sessions, and 27% faster runs by forcing Claude toward the simplest working solution instead of letting it overbuild.
Caveman strips filler from Claude's responses so answers read closer to 'lot word when few word do trick,' cutting both reading time and output token spend.
The new Claude Code Projects feature reuses the name of an older, different 'Projects' tab, so the two are easy to confuse despite working nothing alike.
Claude Code Projects runs one conversation as parallel cloud sessions that split and reassemble work automatically, instead of forcing you to track which separate chat has which context.
BuildPartner.ai can pull live content from Twitter, Instagram, and YouTube directly into a Claude terminal session, something Claude cannot do natively on its own.
A free local markdown viewer like Obsidian earns its place specifically because Claude generates markdown files constantly, not as a replacement for a team workspace tool like Notion.
Anthropic's own Small Business plugin ships skills like a sales pipeline tracker and an inbox manager, useful less as a product and more as a reference for structuring skills inside any organization.
Dictating a request instead of typing it can surface a better idea mid-sentence, and a saved voice history doubles as a record of your own filler-word habits over time.
Takeaway
Eight habits that cut Claude's cost and chaos
COST-CUTTING STACK
None of these tools make Claude smarter, they change how much you spend, how much you type, and how much of the conversation you have to hold in your own head.
01Lavish — plan visually
A rendered, commentable plan beats a chat thread for anything complex, you can leave a note like 'switch tool order' instead of re-typing the whole request.
Planning visually works for both technical builds and non-technical ones, including mocking up a design's look before Claude writes any code for it.
02RTK — stop paying for noise
Claude reads the entire raw output of a command by default, even when only a few lines of it changed or matter.
Filtering what Claude sees after a command runs is a separate lever from filtering what you ask it up front, and it only covers tool output, not your total token spend.
03Caveman + Ponytail — write less
Output tokens cost you twice: the money for the tokens themselves and the time you spend reading a long answer you didn't need.
Constraining Claude toward the simplest working solution, rather than editing an overbuilt one after the fact, is what actually cuts code bloat.
04Claude Code Projects — not what you think
Running multiple efforts as separate chats forces you to remember which thread has which context, a single project that splits and reassembles work removes that bookkeeping.
A feature reusing an old name for an unrelated new capability is a real confusion risk, check what a feature actually does before assuming you already know it.
05BuildPartner.ai — finding the tools
Claude can't natively pull live content from social platforms, a connected tool that fetches current posts closes that gap directly inside your terminal.
Scheduling a single recurring check-in, like a weekly curated recap, is a concrete way to stop the stress of manually monitoring every new AI announcement.
06Obsidian — reading what Claude writes
A free local markdown viewer is worth having specifically because Claude generates markdown files constantly, you need something fast to read them in, not necessarily your team's full workspace tool.
07Anthropic's Small Business plugin
A vendor's own reference implementation of their tool, here Anthropic's skills built for running a business, is a legitimate way to learn patterns you can adapt for your own organization.
08Hex — talking instead of typing
Speaking a request out loud instead of typing it surfaces new ideas mid-sentence that typing tends to suppress.
A spoken record of your own requests doubles as a way to see your filler-word habits and improve how you communicate generally, not just with Claude.
Glossary
Terms worth knowing.
RTK (Rust Token Killer)
A tool that filters a command's output before it reaches Claude's context, stripping unnecessary text so Claude sees closer to 1,800 tokens instead of a full 15,000-token raw dump.
Claude Code Projects
A newly announced Claude feature, distinct from an older tab of the same name, that runs one conversation as multiple parallel cloud sessions and reassembles the results automatically.
Caveman
A Claude plugin that strips filler language from responses so output reads terse and stripped down, cutting both reading time and output tokens.
Ponytail
A Claude plugin that constrains Claude toward the simplest working code instead of letting it default to an overbuilt solution.
Hex
A free, open-source dictation tool that works system-wide on a computer, not just inside a single app, triggered by a hotkey.
“This visual process of planning has personally changed how I build almost everything.”
concise, standalone claim about Lavish that works as a hook on its own→ TikTok hook↗ Tweet quote
02:28
“A big reason for this is that Claude reads way more information than it actually needs.”
names the exact problem RTK solves in one sentence→ IG reel cold open↗ Tweet quote
06:00
“Lot word when few word do trick.”
playful Office/Kevin reference that doubles as the Caveman pitch→ TikTok hook↗ Tweet quote
14:27
“None of these are about which model is the smartest. They're about making the way you work with the model that much better.”
the video's whole thesis in one line→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
story
So free tool number one is Lavish, and this fundamentally changed how I use Claude. See, the people who build the fastest with Claude spend a ton of their time planning what they want Claude to actually build. But when you're planning anything complex, going through the plan in a chat interface is extremely overwhelming.
You'll get walls of text, you'll have to scroll up and down, and you'll have to try and remember what you decided three messages ago and what's still pending. So instead of all that, I use Lavish, which creates a visual document of your plan, and you can add comments directly on it as you iterate through it. you can see a lavish i created while planning this video out about these eight tools so what i was able to do is if i didn't like the order of the tools i could add a comment like switch lavish to tool number one or i could add another comment that's change this tool with this tool and then once i'm done reviewing it i can hit send to agent and it'll update the plan automatically this visual process of planning has personally changed how i build almost everything and clearly this is both for technical and non -technical work but if you are building any sort of app or visual tool it can show you a mock -up of the design before building it which is extremely powerful now i've added a link to lavish below which is where you'll find every plugin that i mentioned in this entire video and to set up any of these just bring it into claude and say download lavish along with the link and i'll take it from there now that's the first of eight free ai tools i'm going through that will help you build 10 times faster and save you money in the process now the next two tools that i'm going to cover are explicitly so that you don't hit your token limits tool number two is rtk which will help you stop
paying for things that you shouldn't be paying for. So RTK stands for Rust Token Killer, which essentially means save you money. So how does it actually do this?
Well, most of you have experienced your AI tools hitting a token limit. And a big reason for this is that Claude reads way more information than it actually needs. So in the ideal world, you'd only share it the text that actually matters.
To put an example to this, let's say you change one page of a 10 page client report and ask Claude to review it. A person would just typically read the page that changed, but often Claude will just look at the entire thing by default.
So the fix for this is to filter what Claude sees. So it only sees the information that matters most. So normally, right?
Claude makes a change, checks its work by running a command and the fully raw output gets dumped into Claude. Let's say that's about 15 ,000 tokens. When you integrate RTK, it's the same change, the same check, but RTK will actually strip all the unnecessary texts and then it'll land in Claude closer to about 1800 tokens.
And now in their documents themselves, they claim a 60 to 90 % token reduction. but that really depends on what you're doing. And this is effective for people who are technical or not technical.
And now to be clear that this only covers what commands send back to Claude. So it won't cut your... total token consumption by 60 to 90%, but this is a big step in the right direction.
Now I've linked this below and you can run a similar command to what we did in tool one to get it all set up. What we're covering here paired with Claude will help you build almost anything, a tool, a site, a whole product. But once it's live, the question actually changes.
It's not, can I build this? It's, is anyone actually going to pay for this? And if so, how can I get more people to pay for it?
Which brings us to today's video sponsor, WAP. When you start a business or want to scale your current business, there are two key questions that you have to answer. answer the first how will you receive payments and there are a number of providers but I chose WAP to run my business through because it is the cheapest way to accept payments and with its flexible checkout options it makes it so that I can create a plan that works for our customers and this is nice and I love the cheap fees but that wasn't the thing that interests me the most the second question is how can I get more people to pay for my service and for me and most people this question isn't easy to answer so personally I get traffic from Instagram YouTube my newsletter all over the place and it's hard to see who's actually converting.
And plus with my old setup, I couldn't see which of those people actually paid. But WAP puts the web traffic and the conversion analytics all in one place. And to set this up, I went to their dashboard.
They had a single Claude prompt that I was able to copy and paste into Claude and it set it all up for me. Now, when someone lands on the site, WAP sees what page they look at and a full history of their interactions and then ties a sale back to the visit. So I can see visitors, page views, and the revenue that came from the site all in one page and spot the exact step.
where people drop. Now getting the data is just part of it, but you have to then act on it, which is where WAP's economic intelligence comes in. What this does is it looks at all my data and tells me the single best thing that I should do next to grow my business.
This can be optimizing a page, launching an ad campaign. And in this case, it actually suggests that I add payment plans to my funnel to help convert more buyers. And I actually went and did that and helped me convert people to paying customers.
So whether you're starting something new or already have a business that you want to scale, sign up for WAP using the link in the description. Going to the next tool, which is Caveman. and ponytail, which is about making Claude write less.
So yes, this actually is a two for one, but there's a good reason to use both and they solve a similar problem. So I made sense to actually group them together. So the problem that these are solving is when Claude sends back a super long answer, that's hard to follow, it's costing you in two ways.
So it's costing you in time, right? Because you have to read the whole response and digest it. And it's also costing you money because that's token spend.
See the last tool RTK that we covered, that was about saving tokens on the input. This will help you save tokens. on Claude's output.
See, Caveman and Ponytail are designed to constrain the output that Claude produces. So going through Ponytail first, this forces Claude to write less code. And it does this by stopping Claude from overbuilding and instead reaching for the simplest thing that works.
And in terms of what this actually impacts, pulling directly from Ponytail's readme, it says write approximately 54 % less code, up to 94%, approximately 20 % cheaper, and 27 % faster. Now, whether or not it hits those baselines, I don't think it really matters. It's about simplifying the code and the output that it produces because Claude often makes things way too complicated.
Now going to the next tool is Caveman. And this is all about sending you a response with fewer words. This does exactly what it sounds like.
It makes Claude answer you like a caveman. So it'll remove all the filler and it'll keep all the substance. Now you guys know that I'll bring in TV show references whenever I think it applies.
And this actually reminds me of an office quote from Kevin. So here's him just talking about how he wants to be more efficient with his words. Why waste time?
lot word when few word do trick now in that clip kevin is literally just saving tokens now before i dive into like a before and after using caveman if you do like when i bring in these types of references let me know in the comments because i personally love doing it i'm just wondering if they resonate so anyway caveman is what forces your ai output to sound like kevin so on the left you can actually see an output before caveman and then on the right is the after you can see it it looks sounds like a caveman which if nothing else it keeps it fun but it does get rid of all the the fluff, makes it more efficient, and it cuts through the noise.
To install both, here is a prompt you can paste straight into Claude. So far, we've been talking about free tools that enhance Claude, but what about the tools that Claude already has inside of it? Going to tool number four, which is Claude code projects.
And this is not what you think it is. So I recently made a video about the optimal Claude setup, and I explicitly said, don't use Claude projects. But now I'm saying use Claude projects.
Well, it's actually two different features. See, Claude recently announced a new feature called Claude Code Projects.
And yes, they're using the same name and I'm going to have to message the Anthropic team about this because frankly, it is a little too confusing, but this is actually a new feature and it is fundamentally different in terms of how it operates. And honestly, this feature alone, I may make an entire video on this because it really is such a powerful way to interact with AI.
So first, what is it? If you've ever worked on multiple things in Claude at the same time, which you probably have, you've likely thought, which chat should I put this in? Or is this supposed to go in this?
one or that one, or you just find yourself constantly switching between conversations and it's just burdensome. So this problem was so profound that Claude actually built a feature to solve it. So here's a tweet, but a project is one conversation with Claude.
It splits the work into threads itself, runs them as parallel cloud sessions, passes context between them, and keeps going when you leave. So in practice, what this means is you talk to one Claude terminal and it hands pieces of work to other Claudes and manages them accordingly. Now to access this feature, it's currently rolling out to pro and max users for free.
And you'll find it in the left side while in cloud code under a projects tab. But if you don't see it yet, it'll likely come up soon. But if you want something similar in the meantime, there is an open source alternative called First Mate.
And it's actually created by Kun Chen who made Lavish, which is the first tool that we went through. Now that is a lot more technical and I won't do a deep dive on it, but it's linked below. And this process and framework is exactly what my team has been using for quite a while as we work directly with companies, helping them build out their AI infrastructure.
and we love it. So we were super excited when we heard that Claude was integrating it directly into their product. At this point, we've touched on tools that help you build faster and also saves you time and money.
But what about identifying what's actually worth building in the first place? This goes to tool number five, buildpartner .ai, the tool that I use to find all of these tools in this video. So buildpartner .ai is actually a tool.
So one of my challenges has always been searching the internet and social media for up -to -date content and specifically Instagram, Twitter, and YouTube. But with In Build Partner, there's a new feature that lets you pull content from these platforms directly from Claude without ever leaving the terminal.
So for example, I could say, using Build Partner, what are the top free AI tools people are talking about on Twitter? And it'll go and look directly on Twitter and pull those tweets for me. And this is something that Claude normally can't do.
And this is really important for me because I've mentioned this on the channel before, but despite making content in this industry every day, the stress that comes from trying to keep up with the latest trends is very real. And I know a lot of the people watch. this feel that so to avoid this i actually use buildpartner every friday to give me a curated recap of the latest news and so this helps me solve a problem that i personally have and it just makes the quality of my life better and if you are new to this channel buildpartner .ai is my tool that i've built and it starts free and then depending on how much you use it it will eventually cost money but because you're watching this video if you use the link below i will 5x the amount of free usage you get because it i love the youtube fam the thing that i want to stress here is find a way to detach yourself from staying up to date with every single new announcement.
It is just so stressful. And this is just a way to solve that problem. Now getting to free tool number six, which is Obsidian, which is about seeing what Claude is writing.
And I've realized that a lot of people on this channel don't fundamentally understand how this works. So Obsidian is a glorified file viewer and it is entirely free. And the reason that I like it is that I'm constantly having Claude create markdown files that I then want to go in and edit myself.
And the best, cheapest, simplest way to read this locally is using a... tool called Obsidian. And if this is still unclear, just think of this like Apple Notes on steroids for your local files.
And to help you orchestrate in your tool stack, this is something I use for my local files and it doesn't replace what I use for my team, which I personally use Notion for. Obsidian is just a way to iterate on my own machine and read through the files. So if you're using Claude to edit files on your computer, I highly recommend this.
To set it up, it's simple. Just go to Obsidian, download it, and then create a vault, which is just a folder on your machine. Tool number seven is Anthropic.
small business plugin. So I recently released a whole video about this on my channel, which I will link at the end of this video, but Anthropic has a plugin specifically designed for small businesses. And this comes with a whole set of skills across different parts of running a business like grow pipeline for your sales, inbox manager for day -to -day operations.
And the skills are nice, but that's not really why I like it. A lot of you watching this channel are business owners or operators and seeing how Anthropic's team thinks about using different skills across different lines of business will stimulate your creativity inside your own organization.
Yes, if you're on this channel, you know about cloud skills and you likely know the basics about it, but it's one of those tools where it's not about knowing, it's about how you actually implement it. And seeing how Anthropic implements it across your organization is extremely helpful. Now, I did make a whole video on this, which will be linked at the end of this.
But before we get to the final tool, which is probably the one I use the most out of all these, you know the drill. This is our anti -slop agreement. Everything in this video is built by humans for humans.
And if that's worth something to you, subscribe, and help this content reach more people. And it is a thank you to everyone who supports.
You guys really are legends. I love seeing you guys in the comments. I give away a Claude Max subscription as a thank you.
So this video's winner is Truth Revealed by SD, who's building a website to extract information from interrogation transcript. I honestly have a couple questions about that. Anyway, to enter this video's giveaway, comment which tool I covered here that interests you the most.
Get in the last tool, which is Hex. Talk to Claude instead of typing. And if you already use a tool like this, pay attention because there's a couple nuances that I don't think you're realizing.
Hex is a open source free alternative to other voice tools like Whisperflow, and it works anywhere on your computer. And so, yes, even if you don't use an external tool within Claude, there is voice mode, which is powerful, but that doesn't help you outside of Claude. So for me, what I do is I have a hotkey where I can double tap control and I'll start recording my voice and I'll paste it in whatever app I'm using.
So if you're still typing in Claude or any other tool, you're living in 2023 and this is free. And it's just worth a try. Frankly, this has changed how I work so much that I actually have a new problem.
I used to love working at coffee shops and now I almost can't because I speak into my computer 99 % of the time and people just start looking at me funny. And I'm just like, what are you looking at? And then I realized that I'm talking to an AI and I get it.
But the speed alone is worth it. Now there are two non -obvious ones that you're likely not considering. The first is this actually helps you think.
So if you've ever started talking to someone and as you talk, you figure out a new concept in your head. That's the power of a stream of consciousness. And by having an audio recording, this allows you to do that.
And often I'll have an idea in my head and then about halfway through, I'll be like, oh wait, I have a new idea that's actually better. And that's what speaking will stimulate. And the second, and this is something I guarantee you're not doing, is it also can serve as a speaking coach.
So what it does is it records what you say and now you actually know how you speak. So for me, I often say like. I use a lot of filler words and I can look back historically and see how I'm improving over time.
This is a meta point, but the better you can speak, the better you can deliver your points to people and that will impact your ability to provide value in whatever industry you're working in. To download this, it is linked below and it is very straightforward to download. Now to recap everything we've covered, if you use all these tools, your wins will start stacking.
You can spend less with RTK, Caveman, and Ponytail. You can plan better using lavish cloud code projects and build partners. And you can learn how Anthropic thinks about skills from their small business plugin.
Then you can just simplify how you work by using Obsidian, using Hex so you can speak into it. And then for payment processing and customer insights, use WAP, which is also today's video sponsor. And you'll notice that none of these are about which model is the smartest.
They're about making the way you work with the model that much better. And one thing that I did cover briefly was Anthropic small business plugin. But that's something that deserves a ton of more attention.
So go check that out by clicking right here. where I break down the 13 new skills that were just released in that plugin that pair perfectly with everything that I covered today. So I'll see you over there.
Peace.
The Hook
The bait, then the rug-pull.
The pitch up front is that none of these eight tools make Claude smarter, they change how much the reader spends, types, and has to hold in their own head while building with it.
CTA Breakdown
How they asked for the click.
VERBAL ASK
03:05product
“Sign up for WAP using the link in the description.”
Mid-roll sponsor read for payment processor Whop, woven into the video's own token-saving theme by framing payments and conversion analytics as the next problem after you've built something with Claude.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A JP Morgan engineer turned $25M startup COO lays out the seven pieces of Claude he'd learn again, in the exact order that compounds instead of the order most people stumble into.
A rapid-fire tour of seventeen ways to use Claude's design tool for real work, from rebuilding old branding to wireframing a product before writing any code.
00:16
00:25
00:38
00:52
01:01
01:08
01:20
01:34
01:45
01:56
02:05
02:30
02:45
02:48
03:03
03:14
03:35
03:52
03:59
04:07
04:21
04:36
04:43
04:54
05:03
05:13
05:27
05:35
05:48
05:58
06:12
06:35
06:40
06:53
07:08
07:20
07:41
07:48
08:00
08:14
08:26
08:39
08:49
08:59
09:12
09:19
09:32
09:44
09:55
10:17
10:39
10:45
11:01
11:13
11:24
11:35
11:46
11:57
12:12
12:31
12:42
12:50
13:04
13:15
13:37
13:48
14:03
14:11
14:22
14:32
21:11
17:42
16:15
12:15
Video of the Day15:12
13:18