Top 10 NEW Open Source Claude Code Tools (May)

CChase AI
컴퓨터/소프트웨어창업/스타트업구직/면접AI/미래기술

Transcript

00:00:00Hundreds of new open source AI projects hit GitHub every single day,
00:00:04yet only the smallest fraction of a percent are actually worth your time.
00:00:08But today I'm going to be highlighting 10 that are almost every single tool we're
00:00:12going to cover today has just come out within the last month.
00:00:15So unless you are as obsessed with this stuff as I am,
00:00:18I promise you're going to get exposed to at least a few new tools.
00:00:22Now the first tool on the list is my favorite because it's one I use literally
00:00:25every single day at this point. And that is the caveman skill.
00:00:28This repo has gotten over 50,000 stars within its first month of release.
00:00:32And the whole idea is this just a skill that we can use inside of Claude code or
00:00:36codex that makes our agent talk like a caveman AKA it's not going to be
00:00:41so damn verbose.
00:00:43So you have some examples here where you have your normal Claude code response
00:00:46where it kind of just goes on and on and on. But if I use caveman, well,
00:00:50it's kind of just going to get to the point.
00:00:51This repo has taken the idea of why say many word when few do trick and just
00:00:56codified it. Now,
00:00:57the cool thing about caveman is that there's levels to it.
00:01:00Like we don't have to go full in the Neanderthal. We can do caveman light,
00:01:03which is what I sit at. We can also do full or we can do ultra. Now,
00:01:07I will say this repo gets a little excited when it comes about how much,
00:01:11how many tokens you're saving. You'll say like, Hey,
00:01:13we're saving like 75% of output tokens.
00:01:15Understand that the way caveman works is it's just going to be changing how many
00:01:20words you see. It doesn't change how it does its thinking.
00:01:22It doesn't change the amount of stuff it's ingesting. So overall,
00:01:26if we take it all together, you're looking at about a 5% or so savings,
00:01:30when it comes to tokens. And I've done a full video on this,
00:01:32and I'll link that above. If you want to do a deep dive. Now,
00:01:35I think the secret when, when it comes to caveman,
00:01:37and I think they kind of bury the lead here is the idea that our large language
00:01:41models might actually do better if they're forced to give more concise responses.
00:01:45And this comes from a March, 2026 paper. It's called brevity constraints,
00:01:49reverse performance hierarchies and language models.
00:01:52And basically the long and short of it is when we have powerful models and we
00:01:57force them to be concise,
00:01:58they're more likely to give us correct answers because they're essentially not
00:02:01going to talk their way into the wrong answer.
00:02:04And it's actually really interesting study.
00:02:06And I highly suggest you take a look at it.
00:02:07So we take those things together where we're going to be saving tokens and I'm
00:02:11potentially getting an actual quality increase.
00:02:13What's not to love about this thing. And it's just a simple skill installing.
00:02:17This is super easy. You can just run the commands here inside the repo,
00:02:20or you can just copy the repo URL, put it inside of Claude code and say, Hey,
00:02:24let's start running caveman. And if you want to do caveman light,
00:02:26just say caveman light. If you want to do ultra do ultra,
00:02:30it's very easy to execute.
00:02:31I'm always a huge fan of these lightweight tools that give us some wins on the
00:02:34margins without any real downside. So if you check out nothing else here,
00:02:38check out caveman. But before we move on to tool number two,
00:02:41a quick word from everybody's favorite sponsor me.
00:02:44So I recently came out with my Claude code masterclass and it is the number one
00:02:48way to go from zero to AI dev,
00:02:50especially if you don't come from a technical background.
00:02:53I update this every single week and we really focus on real life use cases and
00:02:58building upon the foundation of Claude code with things like an entire agentic OS
00:03:03system. So if that's something that you would be interested in,
00:03:06you can find it inside of chase AI.
00:03:08Plus there's a link to that in the pin comment. Now,
00:03:10tool number two is all about memory and knowledge graphs. And that is graphify.
00:03:15Now what graphify is able to do is it reads our files to build a knowledge graph.
00:03:19And because we now give Claude code a clear structure to understand what we're
00:03:23working with,
00:03:24we're able to execute our tasks while using way less tokens per query.
00:03:28They quote 71.5 times fewer tokens per query versus reading raw
00:03:33files. Now, when we talk about knowledge graphs and memory,
00:03:36a lot of us first start to think about things like obsidian, but obsidian,
00:03:40while this does give us a knowledge graph in theory,
00:03:43and that's what we're looking at now, this isn't a true knowledge graph in the
00:03:46sense of like a graph rag system, something like light rag or rag,
00:03:50everything graphify is much closer to that true rag structure than something like
00:03:55obsidian is remember obsidian for all intents and purposes is just a nice
00:03:59interface for us to be able to deal with Markdown files and Markdown files
00:04:03exclusively graphify is multimodal.
00:04:06Now it's not multimodal in the sense that it's going to be ingesting pure video,
00:04:09something like, you know, Google's embedding too,
00:04:12but it is able to look at things like PDFs screenshots, diagrams,
00:04:16and it's able to take videos and then use whisper to pull what it needs out of
00:04:20there. Furthermore, graphify doesn't use embedding.
00:04:23So when we're talking about sort of that spectrum between something like this
00:04:26obsidian and a true rag system, something like light rag,
00:04:29I would say graphify sits somewhere in the middle.
00:04:31And it's something that we can essentially layer obsidian on top of.
00:04:34So if you're someone who loves obsidian,
00:04:36wants a little extra power in terms of what's going on with your memory and your
00:04:40files under the hood, yet you don't want to take the step into some sort of true,
00:04:44rag system with embeddings and everything like that. Well,
00:04:47graphify might be perfect for you and definitely stay tuned for a deeper
00:04:52dive on this particular topic or on a video that might be coming out in the next
00:04:57few days. Now tool number three is when you probably haven't heard of before.
00:05:00It's Claude video just came out last week. We're at 400 stars.
00:05:03And what it does is it gives Claude the ability to watch video.
00:05:07Now what do I mean by that? Because we know Sonnet and Opus can't ingest video.
00:05:12Well, this tool has a pretty clever approach. Once it's given a video,
00:05:16it uses FFmpeg to extract the frames at a particular rate,
00:05:21depending on the length of the video.
00:05:22Obviously if it was 60 FPS and it's a 10 minute video that would cost an insane
00:05:27amount of tokens.
00:05:28So it gives it a default frame budget based on the duration of the video.
00:05:32So a 30 second video would be 30 frames. If it's 10 plus minutes,
00:05:36it would only be a hundred frames. So it gets kind of sparse,
00:05:38but it essentially feeds screenshots to Claude code.
00:05:42It grabs the audio via whisper and it uses those two things in combination to
00:05:47essentially watch videos. Now,
00:05:49I think this is a really useful tool because when it comes to handling videos,
00:05:53there's really only two other pathways right now when it comes to something like
00:05:56Claude code or codex and that's all right,
00:05:58let's just send it off to something like notebook LM and have it figure it out or
00:06:03kind of in that same category is let's invoke Gemini via an API call and just
00:06:08send it that way. This gives us sort of a, you know,
00:06:13different approach where we aren't beholden to Gemini to deal with these videos
00:06:17for us because we're breaking it down via screenshots.
00:06:19Obviously when we talk about longer videos, three minutes plus 10 minutes plus,
00:06:23you're going to run into issues.
00:06:24Just like what are you actually trying to do with these videos?
00:06:26But I think anything that gets us closer to having Claude could being able to
00:06:30handle video is a great tool for us to play with moving forward. Now,
00:06:34tool number four is when I did a video on recently, and that is open design,
00:06:37which is essentially an open source clone of Claude design.
00:06:42You can now use cloud design or something pretty close to it with any sort of
00:06:46coding agents. So you could do this completely locally for free.
00:06:50You don't even have to be on cloud code.
00:06:51They've copied the exact layout of cloud design in terms of being able to create
00:06:55prototypes, slide decks, and added some additional functionality,
00:06:58like also being able to call APIs for image creation and for video creation and
00:07:03open design itself is really built upon four other open source tools.
00:07:07The first one being who wash you design,
00:07:09which is basically another cloud of cloud design,
00:07:12but it's purely inside the terminal, the Guzheng PowerPoint skills.
00:07:17So allowing us to create these power points and then actually extract them
00:07:20properly as well as open code design and then multi-game.
00:07:24So it's taken all four of those added a package of 31 skills and voila,
00:07:28we essentially have local cloud design.
00:07:30So if you're someone who really likes cloud design,
00:07:32especially the graphical user interface portion of it,
00:07:36I highly suggest checking this out.
00:07:37If you've already hit your usage limits for the week. Now,
00:07:40if you're someone who cares about where your tokens are going and how much money
00:07:43you're throwing away every single month on these coding tools,
00:07:46then you are going to like tool number five, which is CodeBurn.
00:07:49CodeBurn tracks token usage costs and performance across 16 AI coding tools and
00:07:54allows us to get a much better look at where tokens AKA our money is going
00:07:59well beyond what, you know,
00:08:01forward slash usage is going to give you inside of cloud code.
00:08:04You can see in this dashboard, it breaks it down by activity, project, model,
00:08:09core tool, shell commands,
00:08:10MCP servers and shows us not only how many tokens we're using,
00:08:13but like the actual dollar amount, which is really important,
00:08:16especially if you're on the API. Now,
00:08:17more importantly than just telling us where our tokens are going and how we're
00:08:20losing money, it gives us ways to fix the problems.
00:08:23It tells us how to optimize our systems. So we stop burning so many tokens.
00:08:28So just like caveman,
00:08:29I think CodeBurn is one of those lightweight tools that is almost pure upside.
00:08:33So definitely take a look at this one. Tool number six is impeccable.
00:08:36Now impeccable came out a couple months ago,
00:08:39but they recently came out with their 3.0 version just last week,
00:08:43which is why I kind of wanted to include it because their updates to impeccable
00:08:46include the ability to actually edit front end designs in a browser.
00:08:51And if you didn't understand by now,
00:08:52impeccable is a tool for front end design impeccable ships with a single skill.
00:08:57Yet that single skill includes 23 different commands that are all about making
00:09:01sure your web pages don't suck.
00:09:03What I like about impeccable is it includes this website where I can actually see
00:09:06what each and every command does.
00:09:08So it shows a before and an after, and you can see, okay,
00:09:11like what will actually happen if I use this skill? Furthermore,
00:09:15it now has a live mode where you can actually bring up your webpage,
00:09:19click on different components and then go through different variations on the
00:09:23browser itself.
00:09:24I actually did a whole deep dive on this and I will link that video above if you
00:09:28want to see this in action.
00:09:29But I think the best part arguably might just be the website and the ability to
00:09:32see all these before and afters and just kind of give you inspiration for like,
00:09:36all right,
00:09:36here's what my AI slop looks like versus what it should look like and seeing the
00:09:41different ways you can make minor adjustments on individual components,
00:09:45but in totality that can really change the way your website looks and feels.
00:09:48And again, this live mode just got released.
00:09:50So if you've used impeccable in the past without it highly suggest you take a
00:09:54look at it again. So sticking with the front end design theme tool,
00:09:58number seven is design extract.
00:10:00Now a big repo that came out a little while ago and I've talked about in the past
00:10:04is awesome design dot MD.
00:10:06Now awesome design dot MD has taken off since it first came out about two months
00:10:11ago. It's up to 70,000 stars. And the idea is,
00:10:14is they give us this repository of all these popular websites, say for example,
00:10:1811 labs.
00:10:19I click on it and I can see essentially an entire breakdown of what their website
00:10:24looks like from an aesthetic point of view. You know, what are the cards,
00:10:27what are the colors, what's the spacing, what's the font, et cetera, et cetera.
00:10:30The problem with awesome design MD is I can only choose from these. I mean,
00:10:35there's a lot to choose from, but I'm limited as to what I can do.
00:10:38Design extract takes it a little bit further because it's essentially allowing us
00:10:43to get the same thing I showed you here inside of design MD,
00:10:47but for any website we want.
00:10:49So we point this design tool at any website we want to use as inspiration,
00:10:52as a foundation for what we are building.
00:10:54And it's going to grab the layout system responsiveness, interaction states,
00:10:57motion, language, component, anatomy, brand voice, on and on and on and on.
00:11:01So we have a comprehensive thing we can then bring into cloud code and build upon
00:11:06with our brand.
00:11:07And it does all this by using a headless browser to actually grab all this
00:11:10information.
00:11:11So it's a bit more than just taking a couple of screenshots and saying, Hey,
00:11:16copy this. So if you're someone who loves this awesome design repo,
00:11:19but wishes there were some more websites on here that you could essentially use.
00:11:23Well, definitely check out design extract.
00:11:26If you've ever thought about using Claude code to help you apply to jobs or get
00:11:30your resume in order, well, you will like this tool.
00:11:32And that is career ops because that's exactly what it does. As they state here,
00:11:36career ops turns any AI coding CLI into a full job search command
00:11:41center. It evaluates the offers of the jobs out there.
00:11:44It generates tailored PDFs. It scans portals of processes in batch,
00:11:48and essentially tracks everything related to the job search process,
00:11:52which is brutal. And importantly,
00:11:53this isn't a tool that's just like a mass application tool. This isn't like, Oh,
00:11:58go on LinkedIn. And now apply to every single job under the sun.
00:12:01Like this is much more of a scalpel.
00:12:02That's going to tune your resume to the job and make sure the jobs you're actually
00:12:07looking at makes sense for you. This isn't just like, all right,
00:12:09go out there and just like throw up all over the job application process under the
00:12:14hood. It's using playwright to actually navigate the pages.
00:12:17It evaluates the fit based on your CV and then adapts it per each listing.
00:12:21And here's how the general flow works. You paste in a job URL or description.
00:12:25It then classifies it and then figures out,
00:12:28are you a match before then generating a report, the PDF,
00:12:32and then updating the tracker. So definitely a useful tool.
00:12:34If you or anyone you know is trying to leverage something like cloud code to help
00:12:38them in their job search. Now tool number nine is one.
00:12:41I think you're going to hear a lot more about, and that is browser harness.
00:12:44So think of playwright if playwright was self-improving after every single run.
00:12:48So the way it kind of works is if I used browser harness to say,
00:12:52do something on Amazon,
00:12:54every time it went to complete a task on Amazon as this autonomous browser agent,
00:12:59it would update its own agent skill file saying, okay,
00:13:02this is what we did for Amazon. Here's what worked here. Didn't it?
00:13:05Almost in a sense, almost like a mini Ralph loop where we've given it a task.
00:13:09It's going to always update its files to see, Hey, did it work? Did it not work?
00:13:13What did we already try? And then try again,
00:13:15based on the information it wrote about itself and sort of the like self healing
00:13:20thing. So it's still pretty new. It's only been out for a couple of weeks.
00:13:23It's just under 10 K stars,
00:13:24but I think this sort of agentic approach to these browser agents is something
00:13:30you're going to see a lot more. Now I cheated on the last tool on the list,
00:13:33because it isn't technically open source.
00:13:35And even N8N itself isn't technically open source. It's fair use, but you know,
00:13:40you can use it locally. So it gets a little confusing. And that is the N8N MCP server.
00:13:44Now I think the death of N8N has been greatly exaggerated, but let's be honest.
00:13:48It isn't in the same place. It was even six months ago yet.
00:13:52They've begun to realize and pivot into being a tool that Claude code can
00:13:57use very, very easily, especially with this brand new MCP server.
00:14:01So this MCP server is a little different than any other N8N MCP server that has
00:14:05come out because there's been a few out there and they were open source.
00:14:09The difference is this one uses TypeScript instead of just trying to generate a
00:14:12JSON file automatically. So I give the N8N MCP,
00:14:16some sort of command like build me whatever automation it then builds it in
00:14:21TypeScript, which allows it to actually validate the automation to see, Hey,
00:14:25do these nodes make sense? Will this actually work from there as a last step,
00:14:30it gets changed to JSON and then it populates inside your instance.
00:14:33So if you're someone like me who still really likes N8N and there are use cases
00:14:38for it, although it can be kind of niche, this is an awesome tool.
00:14:40Just came out a few days ago and I did a full video on that as well.
00:14:43And I'll link that above.
00:14:44So those are my 10 favorite open source tools for Claude code that have come out
00:14:49within the last month or so. Like I said,
00:14:51this space is literally always changing. It is impossible to keep up.
00:14:55So I hope by watching this you were able to see at least a few of them that you
00:14:58might want to check out. As always, let me know what you thought.
00:15:01Make sure to check out chase AI plus if you want to get your hands on that
00:15:04masterclass. And besides that, I'll see you around.

Key Takeaway

Ten new open-source tools for Claude Code focus on reducing token costs through brevity and knowledge graphs while expanding model capabilities to include video processing, front-end design editing, and self-improving browser automation.

Highlights

  • Caveman Skill reduces response verbosity to save approximately 5% in total token costs while potentially increasing model accuracy through brevity constraints.

  • Graphify builds multimodal knowledge graphs from files and videos to achieve 71.5 times fewer tokens per query compared to reading raw files.

  • Claude Video enables video analysis by using FFmpeg to extract a frame budget between 30 and 100 screenshots based on video duration.

  • CodeBurn provides a specialized dashboard to track the actual dollar amount spent on API tokens across 16 different AI coding tools.

  • Design Extract utilizes a headless browser to pull layout systems, interaction states, and brand voice from any live website for use as a design foundation.

  • Browser Harness implements an autonomous browser agent that self-improves by updating its own skill files after every execution on a specific site.

  • The N8N MCP server uses TypeScript to validate automation nodes before converting them to JSON for workflow population.

Timeline

Token Optimization via Caveman Skill

  • Caveman Skill enforces three levels of brevity—light, full, and ultra—to eliminate verbose AI responses.
  • Concise responses from powerful models reverse performance hierarchies and lead to higher accuracy by preventing the model from talking itself into errors.
  • Total token savings sit at roughly 5% because the tool only affects output length rather than input ingestion or internal reasoning.

The tool uses a codified approach to ensure Claude gets straight to the point. A March 2026 paper titled Brevity Constraints Reverse Performance Hierarchies in Language Models supports the idea that forcing conciseness improves correctness. Users can install the skill directly via a repository URL or through Claude Code commands.

Knowledge Graphs and Video Analysis

  • Graphify creates structured knowledge graphs from PDFs, screenshots, and diagrams to minimize query overhead.
  • Claude Video bypasses native model limitations by feeding specific frame screenshots and Whisper-extracted audio to the model.
  • Frame budgets for video analysis scale from 30 frames for short clips to 100 frames for videos exceeding 10 minutes.

Graphify acts as a middle ground between basic Markdown interfaces like Obsidian and complex RAG systems using embeddings. It leverages structured data to reduce token usage by over 70 times compared to raw file reading. Claude Video offers an alternative to Gemini or NotebookLM by breaking videos down into digestible image-based frames.

Open Source Design and Cost Tracking

  • Open Design integrates four distinct open-source projects to replicate Claude's design interface locally and for free.
  • CodeBurn monitors performance and costs across 16 AI coding tools including MCP servers and shell commands.
  • Advanced cost tracking allows users to identify exactly which projects or models are burning the most financial resources.

Open Design combines tools like WhoWashu and Guzheng PowerPoint skills to allow local prototyping and slide deck creation. It includes 31 skills for image and video API calls. CodeBurn provides a more granular dashboard than standard usage commands, offering optimization suggestions to stop unnecessary token waste.

Front-End Design and Competitive Analysis

  • Impeccable 3.0 allows developers to edit front-end designs directly within a live browser environment.
  • Design Extract automates the collection of responsiveness, motion, and component anatomy from any URL.
  • Visual before-and-after comparisons help developers identify and fix low-quality 'AI slop' in web components.

Impeccable features a single skill with 23 commands focused on web aesthetics and functional variations. Design Extract expands on static repositories by using a headless browser to scrape the layout and brand voice of any chosen website. This data provides a comprehensive foundation that can be imported directly into Claude Code for further building.

Job Search Automation and Agentic Browsing

  • Career Ops transforms CLI tools into job search centers that generate tailored PDFs and track application status.
  • Browser Harness utilizes a self-healing loop to update its internal logic based on what worked or failed during previous tasks.
  • The N8N MCP server validates TypeScript-based automations before populating them into a local instance.

Career Ops uses Playwright to navigate job portals and evaluate how well a user's CV matches a specific listing. Browser Harness represents a shift toward agentic tools that learn from their own history on sites like Amazon. The N8N MCP server specifically improves on older versions by ensuring node validity through TypeScript before final JSON generation.

Community Posts

View all posts