Chase walks a fifteen-minute carousel of GitHub repos — caveman, Graphify, Claude-Video, OpenDesign, CodeBurn, Impeccable, design-extract, CareerOps, browser-harness, n8n MCP — that all dropped in the last thirty days.
Ten open-source tools released in the past month extend Claude Code's capabilities in token reduction, knowledge management, video processing, design automation, cost tracking, job search, and self-healing browser agents.
Ten new open-source projects extend Claude Code in ways worth your attention, and the through-line is that small, focused tools deliver more leverage than heavy frameworks. The mechanism is a survey across categories: token economy (Caveman compresses output and may improve accuracy by forcing concision, CodeBurn audits where tokens and dollars actually go), knowledge and ingestion (Graphify builds a multimodal knowledge graph between Obsidian and full RAG, Claude-Video feeds frames plus Whisper audio), front-end craft (OpenDesign clones Claude Design locally, Impeccable polishes UI components, DesignExtract clones any site's design system), and agentic workflows (CareerOps tailors job applications, browser-harness self-heals between runs, n8n's TypeScript-validated MCP server). The practical takeaway is to layer these lightweight skills onto your existing CLI rather than waiting for
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.
Create a free account →
Firehose problem stated, ten-tool promise made, freshness bar set to the last 30 days.
An Anthropic-style skill that makes Claude Code answer like a Neanderthal. Four levels (Lite / Caveman / Full / Ultra). Repo claims ~75% token savings; Chase recalibrates to ~5% in practice because thinking-tokens are untouched. Links the March 2026 arXiv paper 'Brevity Constraints Reverse Performance Hierarchies in Language Models' as the actual win — forced concision raises answer quality.
Quick promo for his own Claude Code masterclass inside Chase AI+, updated weekly, framed for non-technical learners building an 'agentic OS.'
Builds a real knowledge graph from your files. Claims 71.5x fewer tokens per query vs raw-file reads. Sits between Obsidian (just a markdown viewer with a pretty graph) and full graph-RAG systems like LightRAG. Multimodal — ingests PDFs, screenshots, diagrams, and videos via Whisper — but no embeddings.
Gives Claude the ability to 'watch' video. FFmpeg pulls frames at a duration-aware budget (30s = 30 frames, 10min+ = 100 frames cap), Whisper handles audio, both streams feed back to Claude Code. Alternative to round-tripping through NotebookLM or the Gemini API for video.
Open-source clone of Anthropic's Claude Design. Runs against any coding agent, free, fully local-capable. Built on four upstream tools (Huashu Design CLI, Gooszang PowerPoint skills, OpenCodeDesign, MultiCAD) plus 31 packaged skills. Pitched as the escape valve when you've hit your weekly Claude Design quota.
Tracks token usage and cost across 16 different AI coding tools — way beyond what /usage shows in Claude Code. Breaks costs down by activity, project, model, core tool, shell command, MCP server. Critically, it doesn't just measure waste — it suggests optimizations.
One skill, 23 commands, all aimed at killing generic AI front-end output. v3 added a live mode — click components in a browser preview and cycle variations in place. The repo's website ships before/after pairs for every command, doubling as a design lookbook.
Companion to awesomedesign.md but for any URL. Headless browser scrapes layout system, responsiveness, interaction states, motion language, component anatomy, brand voice from a target site, then ships the bundle to Claude Code as a design foundation.
Turns any AI coding CLI into a job-search command center. Playwright navigates job portals, parses descriptions, evaluates fit against your CV, tailors the resume per listing, generates PDFs, and updates a tracker. Explicitly framed as a scalpel — not a LinkedIn easy-apply spammer.
Playwright-style browser automation that rewrites its own skill files after every run. Per-site memory of what worked / what didn't, with a mini-Ralph-loop retry pattern. Just under 10k stars, only a few weeks old.
Anthropic's n8n MCP server. Difference from prior community attempts: it builds automations in TypeScript first, validates that the nodes wire together, then transpiles to JSON for the n8n instance. Closes the 'generated workflow won't import' bug. Released a few days before this video.
Recap and second pitch for Chase AI+ masterclass.
A monthly 'ten tools that dropped in the last 30 days' video is the lowest-friction recurring format in the Claude Code creator economy — and Joe already has the audience for it.
“Why say many word when few do trick.”
“It doesn't change how it does its thinking. It doesn't change the amount of stuff it's ingesting. So overall, you're looking at about a 5% savings.”
“When we have powerful models and we force them to be concise, they're more likely to give us correct answers because they're essentially not going to talk their way into the wrong answer.”
“Obsidian, for all intents and purposes, is just a nice interface for us to be able to deal with markdown files.”
“This is much more of a scalpel that's going to tune your resume to the job — not just throw up all over the job application process.”
“Almost like a mini Ralph loop where we've given it a task. It's gonna always update its files to see, hey, did it work? Did it not work? What did we already try?”
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.
Chase opens with the firehose problem every Claude Code user has — hundreds of new repos a day, almost none worth installing — then promises a curated ten. Almost every tool he covers shipped inside the last month, which is the real bait: this isn't a greatest-hits list, it's a now-list with an implicit expiration date.
From the March 2026 arXiv paper 'Brevity Constraints Reverse Performance Hierarchies in Language Models': forcing a stronger model to be concise raises accuracy because it can't ramble itself into a wrong answer. Token savings are the side effect, not the prize.
Four graded verbosity settings so you can tune how Neanderthal you want the agent to sound — Chase sits at Lite.
Three tiers of personal-knowledge tooling on one axis: Obsidian (markdown viewer with a graph drawing), Graphify (real knowledge graph, no embeddings, layered on top of Obsidian), LightRAG / rag-anything (true graph-RAG with embeddings). Pick the rung that matches how serious your retrieval needs are.
Claude-video defaults: 30s video = 30 frames, 10min+ video = ~100 frames. Token cost is dominated by frames, so the budget scales sub-linearly with duration. Direct parallel to how mod-watch caps at 80 frames hybrid.
OpenDesign is a meta-package — one repo that stitches four existing open-source design-agent projects plus 31 skills into a single Claude Design experience.
Per-task agent skill file that the agent itself rewrites after each run — capturing what worked, what didn't, what's already been tried. A mini Ralph-loop baked into a tool, not a meta-process.
Instead of generating the deploy-target JSON directly, generate validated TypeScript, prove it compiles and the graph wires up, then transpile down. Validation lives in the source language, not in trial-and-error import failures.
“Make sure to check out Chase AI plus if you wanna get your hands on that master class.”
Soft outro pitch back to Chase AI+ — same product he sponsored mid-roll at 2:48. Reinforces the CTA at the moment trust is highest, after delivering the promised ten tools.
00:00
00:02
00:06
00:11
00:13
00:1500:22
00:32
00:42
00:52
01:02
01:12
01:22
01:32
01:42
01:48
01:58
02:07
02:17
02:27
02:37
02:4702:57
03:07
03:1003:15
03:25
03:35
03:45
03:55
04:05
04:15
04:25
04:35
04:45
04:5504:57
05:07
05:17
05:27
05:37
05:47
05:57
06:07
06:17
06:27
06:3706:47
06:51
07:01
07:03
07:13
07:23
07:3007:40
07:50
08:00
08:10
08:20
08:30
08:4008:50
09:00
09:03
09:13
09:23
09:33
09:43
09:5309:55
10:05
10:07
10:17
10:19
10:29
10:30
10:39
10:43
10:4510:49Add 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.
15:13A 15-minute breakdown of the two-part feature that lets Claude spawn hundreds of isolated sub-agents for complex tasks and why the default single-session approach fails.
June 8th
13:57How a plain markdown vault with one index file replaces a vector database for most solo builders.
April 4th
20:07Ten tools the ecosystem built to fix the one thing Claude Code cannot do on its own.
April 15th
16:27A 16-minute argument that Ralph is a weapon — and most builders need the whole armory first.
January 22nd
14:10Claude Code doesn't generate video pixels. It writes a function that returns one frame for any given time, then a headless browser and FFmpeg turn that function into an MP4 for free.
October 6th
05:34Claude Sonnet 5.5 launched today claiming to beat Opus 5.5 on a coding benchmark at half the token cost, so this AI YouTuber pulls up Anthropic's own announcement page and checks the numbers live.
September 28th