I almost paid $4,000 a year to rent my own voice back to me.
ElevenLabs. $330/month.
The model that powers it? Open weights.
The inference? Runs on 3GB of RAM.
The voice cloning? A 30-second sample and a model you can download for free.
The only thing between you and the exact same output is a setup guide nobody bothered to write.
So I wrote one. Then I open-sourced it.
The gap
The gap between what AI costs to run and what people charge you to access it is the biggest arbitrage in tech right now. And it's not in your favor.
Every narration tool, every chatbot seat license, every per-user AI subscription — most of them are wrappers around open models with a billing page bolted on.
The technology is free. The packaging is where the margin lives.
I was paying it too. $330/month for narration I could run on a Mac Mini sitting on my desk.
So we built Ghost Narrator. Self-hosted narration pipeline.
Github link: https://github.com/getsimpledirect/ghost-narrator.
Qwen 3.5 14B for text processing, Qwen TTS for voice cloning.
Same quality. No monthly fee. No API limits. No terms of service that let someone else decide what your voice sounds like.
Took three weeks to get right. Now it runs on a laptop.
The contradiction
I've spent the last year telling people to own their tools instead of renting them.
That the future belongs to people who control their infrastructure.
That every subscription you pay for something you could run yourself is a dependency you're choosing.
You can't say that and then keep your setup scripts in a private repo.
Everything I've ever published is free.
Every essay, every video, every framework. The audience is never the customer. That's been the rule since day one.
But the tools — the pipelines, the configs, the three weeks of figuring out why the audio kept clipping — that was all sitting behind closed doors.
Which meant I was doing the exact thing I was telling other people not to accept.
So starting now, all of it goes public. Every tool, every pipeline, every setup script we build that isn't core to the business. Open source. Free.
Ghost Narrator is the first one. It's live now. If you're paying for AI narration — ElevenLabs, PlayHT, whatever — you don't have to anymore.
Why
The business isn't the tools. It never was.
The business is the judgment underneath — knowing what to build, what to kill, and what to own versus rent. The tools are how I show you that's possible.
But there's a more honest answer.
I spent years building things that AI eventually ate. Products, services, information businesses — all compressible, all replaceable.
The thing that survived was me. The identity. The relationships. The willingness to take a position and stand behind it.
None of that lives in a private repo. None of it gets weaker because someone else has access to my narration pipeline.
If I keep the tools locked up, I'm doing what ElevenLabs does — packaging something free and charging for access.
The only difference is I'd be charging attention instead of money.
The real cost of convenience
Every month you pay $330 for narration you could run locally, that's $330 worth of dependency.
Every seat license for a chatbot running on an open model is another piece of infrastructure you don't control, hosted on someone else's servers, in someone else's country, under someone else's laws.
I'm not asking you to care about this on principle. I'm asking you to look at the bill.
There's no Discord. No community. No foundation. We're three people releasing our internal tools as we go. The first repo is up. More are coming.
If they help you, good. If you fork them and build something better, that's the whole point.
— George