jaredrhod

Nine minutes shows you everything below, working. Watch it, then install it.

"Jarvis" is just my name for an AI setup with three parts. You pick which ones you want.

• Memory. A folder of plain text files on your computer. Your AI reads them at the start of every conversation and writes to them as you work, so it never forgets your business again. This is the AI Memory Vault from my videos.
• Voice. Hold one key, talk out loud, let go. It answers through your speakers about a second later in a real voice. Same AI, same memory, no typing.
• Face. A full screen page that animates while your AI works, so you can see at a glance whether it is listening, thinking, or answering. Four designs, including the circuit board from the videos.

Optional extra: Hands. Control elements on your screen with your bare hands. No goggles, no gloves, no headset. It's like virtual reality without the hardware. It opens in its own window when you want it, instead of the visualizer. Add it with the others or come back for it later; same command either way.

Only want the memory system? Same command. Skip the rest.

• 1.Install Claude Code. Not the Claude app, not the website: Claude Code lives in your terminal and can actually build things on your computer. If you've never opened a terminal and typed claude, you need this step. You'll also need a Claude account on a paid plan (Pro works). Open Terminal (press Cmd+Space, type "terminal", press Return) and paste this:

copy the Claude Code install

curl -fsSL https://claude.ai/install.sh | bash && { grep -q '/.local/bin' ~/.zshrc 2>/dev/null || echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc; } && export PATH="$HOME/.local/bin:$PATH" && echo "All set. Any PATH note above is already handled. Type claude to start."

When the install finishes, it just says thanks. Nothing opens. That's normal, you're not done yet.

Note: Mac stopped with "xcrun: error: invalid active developer path"? That's your Mac missing Apple's Command Line Tools. Run xcode-select --install, click Install on the popup, wait the few minutes it takes, then paste the command above again.

• 2.Type claude, press enter, and walk its quick setup. Pick a text style; when it asks how to sign in, choose the first option, "Claude account with subscription" (not "Anthropic Console account", that's pay-per-use developer billing), and log in when your browser opens. If it asks whether you trust the files in this folder, say yes. Then type /exit. Already have Claude Code signed in? Skip to step 3.

• 3.Signed in? Paste this in the same terminal window:

copy the install command

mkdir -p ~/my-agent && cd ~/my-agent && git clone https://github.com/jaredrhod/fullstack-agent && cd fullstack-agent && claude "set me up"

Claude Code opens with the installer already talking to you. It asks your name, which pieces you want (the memory, the voice, the face, and the hands if you want them), and builds what you picked. Take everything and your screen ends up a living circuit board that speaks first: "Hello [you], what are we working on today?" Take just the memory, and you've got the vault system from my videos, built in one conversation.

Already built some of this before, from my videos or prompts? Same command. The wizard finds what you have, keeps what's yours, and upgrades the rest. It never deletes anything you made.

Watch the series first. You need to know this stuff as well as your agent does, or the files below won't help either of you. This is what I actually do to make money in my own businesses:

Then give it all to your AI. The same playbook as files your agent reads before it writes a word. This one doesn't need Claude Code:

Have Claude Code? Paste this; your agent runs the setup wizard and installs it into your vault, as a skill, or both:

copy the prompt, paste it into claude code

I'd like to set this up, please: https://github.com/jaredrhod/ai-marketing-skills.git

Just regular Claude? Download the ready-made skill and upload it to Claude as a skill. Same playbook, no terminal.

"I don't want a Jarvis. I just want the memory system." They're the same system: a Jarvis is the memory system with a voice and a face added on. Run the install in get-a-jarvis, and when it asks which pieces you want, take only the memory. You get the exact vault system from my videos.

"My AI says it can't run things on my computer." You're talking to the Claude app or the claude.ai website, not Claude Code. They're different things. The tell: if you've never opened a terminal and typed claude, you don't have Claude Code yet. /start fixes that in about two minutes.

"Do I have to use Claude Code?" Claude Code in the terminal is the path: it's what everything here is built and tested on, it's what I teach, and it's what the community supports. The honest asterisk: the memory system, the marketing playbook, the rules, barehands, and the visualizer are plain files, so other terminal agents (Codex, Grok, Gemini CLI, a local model) can drive them with rough edges. Two things genuinely need Claude Code: the voice (built on Claude's agent SDK) and the one-command installer.

"What runs on what?" The quick map:
- The memory vault: built and tested on Claude Code. Other terminal AIs (Codex, Grok, Gemini CLI, local models) drive it with rough edges.
- The voice (backtalk): Claude Code ONLY. It's built on Claude's agent SDK.
- The face (ai-visualizer): works with any AI, nothing Claude-locked. It's your agent's face, and it pairs with the voice for the live show.
- The hands (barehands): works with any AI. Any program that can write a file can be its brain.
- The marketing skills: run everywhere, including regular Claude with no terminal (the skill zip).
- The installer (fullstack-agent): Claude Code ONLY.
And the plan question: the $20 Pro plan is enough for all of it. The free plan doesn't include Claude Code.

"Can I run this with Hermes, OpenClaw, or an Obsidian AI plugin?" I don't recommend it and we don't support it. Those put a layer between you and your agent, and that layer is where these setups break. My system is deliberately simple: an agent in a terminal reading your files directly. Obsidian itself is part of the setup (you need it), and MCPs are fine (those are tools your agent calls). Harnesses and orchestrators are not.

"Do I actually need Obsidian?" Yes. It's part of the setup, not an add-on. Your agent's memory is plain markdown files, and the system technically runs without the app, which is exactly how people get talked out of installing it. But Obsidian is how YOU see, browse, and own that memory; a brain you can never open isn't the product. It's free, the installer sets it up for you, and if your setup somehow finished without it, tell your agent: "install Obsidian and connect it to my vault." (Obsidian SYNC, the paid sync add-on, is a different thing and you don't need it.)

"Do I need a paid Claude plan?" Yes. Pro works. The free plan doesn't include Claude Code.

"Do I need Obsidian or git installed first?" No. The installer checks for both and installs them for you, with your OK, as part of setup.

"Will installing overwrite what I already built?" No. The wizard finds what you have, keeps what's yours, and upgrades the rest. It never deletes anything you made.

"I typed claude and it says command not found." Mac and Windows both: the installer puts the claude program in a folder your terminal doesn't check out of the box, then leaves fixing that to you. The install command on this page already includes the fix, so the quickest cure is to run step 1 again with your system's tab selected, it's safe to re-run. Then type claude again. On Mac the standalone fix is: echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc and it prints nothing when it works; silence is success.

"I ran the install command and Claude Code acts like nothing's there." The download step failed before Claude Code opened. Type /exit, scroll up, and read the error message sitting right above where Claude started. Fix what it says, then paste the command again.

"I closed the window in the middle of setup." Nothing's lost. Open a new terminal, go back to the my-agent/fullstack-agent folder, run claude --continue, and tell it "we got cut off, keep going with the setup."

"My agent forgot everything." Three causes, in order of likelihood. One: you launched Claude Code outside your agent's home folder. The boot file only loads from there, so you got a stranger. Use the Chat desktop icon, or open the terminal in the home folder. Two: you wanted the old CONVERSATION back, and a fresh session doesn't replay it. Run claude --continue in the home folder and the last session reopens mid-thought. Three: the vault isn't wired in. Tell your agent: "check that my CLAUDE.md points at my vault and that you read it at startup; fix anything wrong."

"How do I pick up where I left off?" claude --continue, run in your agent's home folder, reopens your most recent session with its short-term memory intact. Closing a window never loses anything. And when you're done for the day, say so: your agent offers to write the daily note.

"Can I use your stuff in my business?" Yes. Everything is free to use commercially. The one thing you can't do is take the code, re-skin it, and sell it closed-source.

"Can one agent run multiple businesses?" Yes, that's how I run mine: one vault, one root folder per business, one agent. Everything Company A lives in the Company A folder, so nothing bleeds into Company B. Open a session and say which business you're on; want two going at once, open a second terminal window. Same agent, same personality. Structure does the separating.

"Do the marketing skills need Claude Code?" No. Download the ready-made skill zip from the ai-marketing folder above and upload it to regular Claude. Claude Code just makes the install nicer.

"I already have my own agent. Can I add just the voice or the visualizer?" Yes. Say one sentence to YOUR agent, in its own session; each repo's README carries the sentence, and /prompts lists them.

"Does the visualizer work without the voice?" On its own it runs demo mode, which is pretty. It performs your real conversations when a voice is wired in.

"Can I have Jarvis's personality?" Yes. The boot template in ai-memory-vault ships with him already in it. Keep him, rename him, or replace him.

"Does this work on Windows?" Fully. Every command on this site has a windows toggle for the PowerShell version, and the installer needs no git up front.

"Can my agent do [whatever you're thinking]?" Probably. Give it the goal, not the task: say what you're trying to accomplish and why, with as much context as you can, and ask it to propose ways to get there. Dry-run it, tell it what was wrong and why, run it again. When it works, have it log the method in Obsidian so it never gets figured out twice, and if you want it fully hands-off, ask it to turn the method into a scheduled automation (a cron job), which doesn't even spend tokens once it's running. If the job needs a tool plugged in, your agent will say so and offer to wire one.

"Something's broken. Where do I get help?" First move, always: ask your agent to fix it. Every repo ships a troubleshooting guide written for your agent to read; fixing this stack is its job, not yours. Still stuck? The Discord, right above this.

It all ties together: a Jarvis is built from my memory system, and the memory system uses Obsidian as your agent's brain. Learning Claude Code and Obsidian IS learning how the Jarvis brain works. And the marketing series teaches you to build sales funnels, then teach your AI to build them too.

Neat trick: YouTube transcribes every video automatically, so Claude Code can "watch" any of my videos by reading the transcript. Just give it the link and ask.

Want me involved directly? Fill out the form and tell me what you need.