Not sure what to fix first?
Most people don’t need another AI tool. They need to know which thing is actually costing them the most.
The AI Bottleneck Finder asks you a handful of questions, then tells you the one constraint capping your revenue right now plus two builds you could ship this week. Five minutes, free, no call required.
Take the free AI Bottleneck Finder →
If you’ve seen any of the Marvel movies, you already know what JARVIS is. The question is how close we actually are to a real one doing serious work for you.
Closer than you think. You can make JARVIS today with a free GitHub repo, and it plugs straight into Claude Code. That means it gets every skill, every MCP connector and every rule you already built. You talk into your mic, it takes the action, and the answer shows up on screen.
I’m Charles J Dove, and I run Charlie Automates. I set this up the morning I found the repo, then tested it on the 3 jobs business owners ask me to automate most. Here’s the full setup and what happened.
What Is This JARVIS Setup, Exactly?
This JARVIS setup is a desktop app that hosts Claude Code through the Agent SDK and gives it a voice, a face and a screen. Claude Code underneath is the same engine you run in your terminal, with every workspace tool attached.
So JARVIS isn’t a new AI. It’s a new front door to the one you already trained.
When I asked my JARVIS to explain what I do in 2 sentences, it pulled that straight from my own files. When I asked how it connects to my second brain, it explained BASE, the local CLI that hooks into Claude Code at session start, on every prompt and before every tool call. Nothing is remembered by the model. It’s recalled from the graph.
That’s the part people miss. The voice is the shiny bit. The value is that it already knows your business.
How Do You Make JARVIS in 2 Steps?
Making JARVIS takes 2 steps: install the repo, then connect a voice model. The whole thing runs from one Claude Code session.
Step 1: Install JARVIS. Open the GitHub repo, click the Code button and copy the repo link. Then open Claude Code and tell it to install JARVIS and boot it locally, and paste the link. It opens in your browser, and you click or speak to activate it.
Step 2: Connect the voice. Out of the box JARVIS can’t talk yet. You have 2 solid cloud options:
| Voice option | Why pick it | Setup |
|---|---|---|
| Fish Audio | Has a JARVIS-style voice ready to go, cheap API usage | Developer tools, create an API key, paste it into Claude Code |
| ElevenLabs | The one most people already know, custom voice cloning | Same pattern: create a key, hand it to Claude Code |
In the same session you used to install JARVIS, tell Claude Code to use your Fish Audio API key and find a JARVIS voice. Restart JARVIS and it talks back.
One upgrade I’d do immediately: add the Fish Audio MCP server to Claude Code too, using that same key. That gives Claude more arms inside the platform, like cloning voices or searching for the exact voice you want.
Why Give JARVIS Full Access?
Full access matters because JARVIS runs headless Claude Code sessions, and by default those can only read some of your setup. Without it, you get a talking assistant that can’t actually touch your tools.
The fix is one prompt back in your terminal. I told Claude Code to give JARVIS full admin access, then attach my rule system: BASE, my CLAUDE.md, my hooks and my memory folder. The goal is simple. JARVIS should be the same Claude Code you use in the terminal, just with a voice.
My setup shows every connection it can reach: Gmail and the Google Workspace CLI for triage, GoHighLevel for my pipeline, Skool for my community, Read.ai and Slack for calls and clients, plus my analytics and creative stack on top.
The resource page with the exact prompts is on the JARVIS toolkit page.
Is JARVIS Actually Practical for a Business?
JARVIS is practical for the boring, repeatable jobs, and I tested it on the ones owners actually want gone. I pulled those from my AI Bottleneck report, which over 300 business owners have filled out. The top asks were email triage, pipeline management, invoicing and building apps.
I skipped app building for this test and ran the first 3.
Email Triage
I asked JARVIS to look through the last 24 hours of Gmail, find what’s important, and draft replies 5 threads at a time with context on top of each one.
It found 5 threads worth answering out of 48 across both of my inboxes. I told it to skip the first 4, create a task in my BASE brain for one client issue, and send that client a short reply saying I’d fix it within 24 hours. It sent it.
Pipeline Management
I asked JARVIS to review the New Report stage in my GoHighLevel pipeline and tell me who’s worth calling.
It came back with 47 real leads plus 8 of my own test contacts. It flagged 14 worth an email and roughly 30 pre-revenue builders who hadn’t started yet. Then I had it write an Apple Note with 5 names to call today, the 9 warm emails, and the test contacts to clear out.
Invoicing
I already have an invoice skill with my brand kit, so I told JARVIS to read a client’s email thread through the GWS CLI and build the invoice without asking me questions. It followed the skill exactly and dropped the PDF in my Downloads folder. Same output I’d get from the terminal.
JARVIS vs Your Terminal: What Changes?
The difference between JARVIS and the terminal is who does the typing. The work underneath is identical.
| Claude Code terminal | JARVIS | |
|---|---|---|
| Engine | Claude Code | Same Claude Code |
| Skills, MCPs, rules | Yes | Yes, with full access |
| How you ask | Type a prompt | Speak it |
| Where the answer lands | Terminal output | On screen, read back to you |
| Best moment to use it | Deep build sessions | Between calls, mid-task, hands busy |
For an owner, that last row is the whole pitch. You’re on the phone all day. You don’t want to open a terminal and re-explain context. You say it, the work runs, and the answer’s waiting.
Is it perfect? No. There’s no such thing as a perfect system, and whether it fits you depends on your use cases. But for triage, pipeline reviews and invoices, it earned its spot.
Key Takeaways
- JARVIS is a front end, not a new model. It hosts Claude Code through the Agent SDK, so it inherits everything you already built.
- Setup is 2 steps: install the repo, then connect Fish Audio or ElevenLabs.
- Full access is not optional. Headless sessions can only read part of your setup until you grant admin access and attach your rules, hooks and memory.
- It found 5 threads worth answering out of 48 and sent a real client reply by voice.
- It sorted 47 leads into 14 emails and 5 calls today and wrote the call list to Apple Notes.
- Your skills carry over. My invoice skill ran through JARVIS exactly like it runs in the terminal.
JARVIS is one piece of a bigger build. If you want the architecture it sits inside, read the 3-step system to build an agentic OS.
Where to Go From Here
Find the one bottleneck killing your revenue.
30-minute free call with Charles. We diagnose your highest-leverage AI bottleneck, install Charlie OS on your machine in one hour, and map the path forward on the same call.
Want the JARVIS files and prompts? They’re on the JARVIS resource page on charlieautomates.com, and the full walkthrough lives in the classroom when you join CC Strategic AI on Skool. The community is free, with a starter module that teaches you how to set up Claude Code from zero.
At my agency CC Strategic, we build systems like this for business owners, and new Charlie Automates builds drop every week on my YouTube channel @charlieautomates.
FAQ
How do I make my own JARVIS AI assistant? Install the JARVIS repo from GitHub by pasting the repo link into Claude Code and asking it to boot the app locally. Then connect a voice model like Fish Audio or ElevenLabs with an API key. Restart JARVIS and it talks back.
Is the JARVIS setup free? The JARVIS repo is free on GitHub. The voice runs on a cloud voice API like Fish Audio, which is cheap to use, and Claude Code needs its own subscription. A local voice model is also possible if you want to cut API usage.
Does JARVIS work with my existing Claude Code skills and MCPs? Yes. JARVIS hosts Claude Code through the Agent SDK, so it uses the same skills and MCP connectors. Grant it full admin access and attach your rules, hooks and memory, or its headless sessions can only read part of your setup.
What can a JARVIS assistant do for a business owner? In my test it triaged 48 email threads down to 5 worth answering, sorted 47 pipeline leads into a call list, and built a branded invoice from an email thread. All 3 were done by voice without opening a terminal.
Is JARVIS actually AGI? No. JARVIS is Claude Code with a voice and a screen. It feels like a leap because it already knows your business through your files, rules and connectors, but the intelligence is the same model you use in the terminal.
Which voice is better for JARVIS, Fish Audio or ElevenLabs? Fish Audio is the fastest path because it already has a JARVIS-style voice and cheap API pricing. ElevenLabs is a strong choice if you already use it or want to clone a custom voice. Both connect the same way through an API key.