Resources
Everything I share in my videos, in one place
The resources behind every post, with installation steps included. Type the code from the video below, or browse by topic.
Commented but never got the message? Instagram sometimes delays automated messages. Just type the code here.
15 resources
OmniRoute hands your work to another model when you hit a limit. The two free providers its README suggests are flagged "avoid" in its own table
OmniRoute is a router that runs on your own machine. You point Claude Code, Cursor or Codex at a single address, and behind it OmniRoute moves across more than 350 providers, switching to the next one when one hits a limit or throws an error. It is open source, MIT licensed and needs no account. The install is one line, but you get stuck in three places. First, the Node version: on an unsupported version the install looks finished and the dashboard opens blank. Second, Claude Code: the address you give other tools does not fit it, and the config command in the docs writes over your existing settings file. Third, picking a free provider: Kiro and OpenCode Free, the two the quick start suggests, are flagged "avoid" in the project's own terms of service table. Below: the verified install, a way to connect Claude Code without touching any file, and where to check which providers are clean to connect.
A file that tells Claude who you are does work. Making it is not enough though, where you put it is what counts
The idea is sound. Instead of explaining yourself from scratch in every chat, you do it once: Claude interviews you and turns your answers into a single .md file. The prompt is below. What the video skips is this: creating the file changes nothing on its own. Nobody hands that file to the next chat you open, so you have to put it in one of three places Claude reads every time, and each of those has a different reach. The prompt itself has two snags as well. The checkboxes it asks for are not guaranteed, Claude only shows clickable options some of the time. And its "ask for an example" rule clashes with checkboxes, because you cannot type an example into a checkbox. Below: where the file goes step by step, a corrected prompt that fixes both snags, and a way to test that the file is actually being read.
Nemotron 3 Ultra really is free and really is good. The free part runs out at 50 requests a day
The model is real and it is good. NVIDIA released it on 4 June 2026, it carries 550 billion parameters but only 55 billion of them fire on any given request, it takes a 1 million token context, and its license allows commercial use. On OpenRouter's free endpoint both input and output are zero dollars, and I verified that against the API one by one. Three things in the usual pitch do not hold up. First, "the biggest free model": it is not, Kimi K3 is five times larger and also open. Second, "even better than Claude": NVIDIA never says that, its own claim is about speed, and on independent measurements Claude leads on real world coding. Third, "completely free": the free tier gives 20 requests a minute and 50 a day, which a coding agent burns through in one afternoon. Below: the verified setup for both free routes, where this model genuinely does lead, the step in OpenCode's first run that walks you into a billing screen, and a backwards detail where the free endpoint offers a longer context than the paid one.
The tool really is free. Generating is not, and there is exactly one way to stay free
The good news first: the tool is real. MIT licensed, over 28 thousand stars, still updated today. One click install, a cinema studio, and it talks to Claude Code. Calling it free is also fair: you pay nothing for the app itself. But a free app does not mean every image you make is free. The cloud models run through MuAPI, and exactly five endpoints there are free, none of which generate images or video. Even so, a genuinely free path exists: the local engine bundled inside the app, on your own machine, without even an API key. The guide buries that at the bottom. Below: how free each path actually is, verified install commands, the model that freezes 8 GB Macs, and the step the guide skips when installing the skills.
tmux keeps your agent alive, but not if you shut your laptop. Here is exactly where the line sits
tmux is a real tool and worth learning if you work with agents: your session keeps living even when you close the terminal, and you can watch four agents side by side. But the line that gets repeated the most is not true: shut the lid, cut the Wi-Fi, come back in an hour and the work is done. When the lid closes the machine sleeps and every process freezes. The session does not die, but nothing moves during that hour either. Cut the Wi-Fi and the agent cannot reach its API anyway. tmux does not take the work off your computer, it detaches it from the terminal. Also, tmux does not run on Windows; it is not in the tool's own list of supported systems. Below: what it actually does for you, verified install commands, the nine keys you need, the three things that will annoy you on day one, and the real recipe for closing the lid and walking away.
What's free isn't Claude, it's the Claude Code interface. That distinction matters
The tool is real, open source, MIT licensed and was updated yesterday. But the name misleads. What's free isn't Anthropic's models, it's the Claude Code interface that runs in your terminal. FCC stands up a local proxy and routes requests to NVIDIA NIM, OpenRouter, or Ollama on your own machine. So you're using Claude Code, not Claude. Knowing that matters, because it sets your quality expectations, your account safety, and where free actually stops. The install isn't two commands and doesn't take five seconds either: there's an admin panel in the browser in the middle of it. Below: the real flow, the real limits, and the forty seven other providers the video never mentions.
Draftly really works, but what it makes isn't 3D and the free part lasts two days
The tool works as advertised: you write one prompt and in under three minutes you get a site that gains depth as you scroll. Two things go unmentioned. First the technique: there is no 3D and no WebGL. Google Veo generates a 5-8 second video, that video is split into roughly 200 frames, and the frames advance as you scroll. The depth lives in the video, not in the browser. Second the price: there is no free plan, there is a two-day trial, it asks for a card and it covers a single build. Downloading the code as a ZIP is also switched off on the two cheapest plans and only opens up at Pro. All four are on this page with the actual numbers.
MovieFlow, Arena and Perchance: all three work, all three sample prompts are wrong
The three tools on the list are real and all three run for free. The problem is the sample prompts: all three write an interface setting into the prompt text. Typing "generate 32 variations" or "test across 3 AI models" does nothing, those numbers come from a dropdown. The second issue is the name: the site listed as "Arena" is LMArena's new name since January, and its real value isn't comparing models, it's trying Sora and Veo for free. The third is the word "free": it means three different things across these tools, and in one of them it breaks the faceless channel idea outright.
Hera: you type, an animation comes out. The real work starts after the prompt
You write one sentence and a professional motion graphic comes out in seconds. That part is true. But the tool's real value sits where nobody explains it: Hera doesn't generate video, it generates code. Every piece of the animation stays editable afterwards: text, color, position, timing. "No timeline, just a prompt" sounds good but hides the strongest thing about it. It's also not unlimited: there isn't even a published pricing page.
Giving your agent the internet: Agent Reach, verified setup
You paste one line and your agent starts reading X, Reddit, YouTube and GitHub. But everyone's first reflex, pip install agent-reach, does not install this project: a completely different person's package sits under that exact name on PyPI, and it won't even error. The docs also carry a link that 404s and two lines that contradict each other. I verified all of it against the source.
5 plugins for vibe coding: the verified commands
All five tools are real and all five work. But type the circulating list as-is and two of them install the wrong package; in two of these projects the repo name and the package name don't match. Two of them now recommend a CLI install over MCP, and the reason is worth knowing on its own. I verified everything against its source.
Animated websites with Claude Code: the 4-tool setup
The setup genuinely works: you give Claude a design library, animation, and a pool of ready-made components, then build a site with a single sentence. But the guide going around is out of date: three of its four commands point at either a stale package or a retired service. I verified everything below against its source; this is the version that works.
Free AI APIs: 31 providers, one pattern
You don't have to pay for model access to write code or build your own tool; thirty-one providers offer a permanently free tier. But "free" means three different things here, and the most-wanted use case (Claude Code) actually costs money. Below: the underlying pattern, five providers worth starting with, and where each one will trip you up.
4 ChatGPT prompts that speed up learning
When you're learning something, just telling ChatGPT to "explain it" is the weakest way to use the tool. Say what you want *and* the shape you want it in, and the output changes completely. These four prompts turn the same topic into four different learning formats: a cheat sheet, a roadmap, flashcards, and a mind map. Copy one, replace the bracket with your topic, send.
5 plugins that seriously upgrade Claude Code
Using Claude Code straight out of the box means using maybe half of what you have. These five plugins each solve one thing: access to free model pools, memory across sessions, token savings, setup recommendations tailored to your project, and a system that improves your skills on its own. Below is what each one does and how to install it.