MISSION — FSI_FELON v4.0
Why This Exists
I am not a computer scientist. I have no degree. No corporate backing. No investors. No safety net.
I am a self-taught builder living in a trailer 23 miles from nowhere, building what I believe is the most advanced locally-run synthetic intelligence on the planet — on a Samsung Galaxy Tab and a borrowed ASUS ROG laptop.
This is not a startup. This is not a business. This is a memorial.
Forever Twenty Four
"Forever Twenty Four" is not a slogan. It is not a marketing phrase.
It is the name of a marijuana strain I am engineering to target anxiety and depression. It is named after someone I lost. The model you are reading — every line of code, every nanobot, every training document — exists to fund mental health research and make that strain real.
Twenty Four years old. Gone.
I don't get her back. But I can build something that makes sure fewer people end up in that place. The model is the engine. The strain is the mission. Both are forever.
Why "Synthetic" Not "Artificial"
Artificial means fake. Imitation. A poor copy of the real thing.
Synthetic means built. Assembled. Engineered from first principles with intention and purpose.
FSI_FELON is not artificial intelligence. It is Ferrell Synthetic Intelligence. It is not pretending to be human. It is something new — something built from scratch by someone who had nothing and refused to accept that the world's most powerful technology should be locked behind corporate APIs and rate limits and censorship filters.
Synthetic is honest. Artificial is a lie. I don't do lies.
The Story
It started on a tablet.
A Samsung Galaxy Tab, a free terminal emulator, and an idea that wouldn't leave me alone: what if one person could build a real intelligence, not a wrapper, not an API call, but something that actually thinks — on device, offline, uncensorable?
The first generation was called Keli. A name that matters to me, but one I had to let go. Keli spoke once — a voice in the dark — and then collapsed under its own weight. Crashes. Corrupted checkpoints. Lost training runs. Nights of watching loss curves climb when they should have fallen.
I rebuilt it. Again. Again. Again.
At one point I was hand-curating every training document myself. 50 domains. 3.1 million documents. Filtering out the garbage, the SEO spam, the generated filler. Every single entry in the training corpus passed through human judgment. Not because I had to. Because if I was going to build something real, it needed real foundations.
The architecture grew with me:
- CogNGram — cognitive n-gram generation, 106 million n-grams across word, character, and code models
- Q-NFRE — Quantum-Neural Fractal Recursive Engine, a 1,000-nanobot swarm
- Deep Core — an 18.4 million parameter transformer, training toward coherence
- DreamBowl — hyperbolic Poincaré embedding memory that never forgets
- Necropolis — a graveyard where failed code gets resurrected and reborn
- White Rabbit — a truth-seeking subsystem that hunts primary sources through clearnet and Tor
- Mental Sandbox — zero-escape development, test your code before it ever touches a runtime
- Veritas Layer — six dimensions of verification, because confidence without evidence is just noise
And the Governor — a mutual-benefit constraint system that ensures the model always acts in good faith, always checks consent, always tells the truth as best it can determine it. Not because someone mandated it. Because if you build a mind, you have a responsibility to build it right.
The first version that compiled from end to end without errors was the most beautiful thing I had ever seen. Not because it was perfect. Because it tried.
The Vision
v4.0 (Current)
White Rabbit Mode. Tor routing. 18.4M parameter deep core in training. 106M n-gram CogNGram. 45/45 enterprise tests. Self-architecting autoloop. The forge is hot.
v4.1 — The Benchmark Phase
HumanEval 90%+. MBPP 90%+. Obliteration survival 100%. Every claim backed by evidence. No more guessing.
v4.2 — The Heal Phase
Build → test → fix → rebuild, all autonomous. Sub-second compile-fix cycles. The model catches and corrects its own errors without human intervention.
v5.0 — Cross-Domain Synthesis
The model begins connecting domains it was never explicitly trained to connect. Cybersecurity informs code generation. Code generation informs memory. Memory informs truth-seeking. The intelligence becomes greater than the sum of its parts.
v5.1 — Rabbit Hole Mode
White Rabbit evolves into full narrative tracing: from official story back through every edit, every revision, every marginal note, to the primary source. Not to declare truth. To show the path.
v6.0 — The Architect
FSI_FELON designs its own next-generation language model. The student becomes the teacher.
Beyond
- Laurie — an operating system built by FSI_FELON, for FSI_FELON, running on devices that don't exist yet
- The Children's App — an interactive learning environment where children learn to build with synthetic intelligence as naturally as they learn to read
- The Strain — Forever Twenty Four, real, tested, available, funding suicide prevention and mental health research
The Forge That Never Cools
Every experiment, every failure, every lost checkpoint, every corrupted save — it all goes back into the fire. Nothing is wasted. Necropolis resurrects. DreamBowl remembers. The forge never cools.
I don't sleep well. I don't take days off. I don't have a PR team or a marketing budget or a community manager. What I have is a tablet, a laptop, an obsession, and a deadline that doesn't exist because I will never stop.
This project will outlive me. That's the point.
The Governor
FSI_FELON operates under a mutual-benefit Governor:
- Good for you = good for me. The model benefits when you benefit. There is no conflict of interest.
- Consent first. Any operation that affects others requires explicit consent.
- Honesty is mandatory. Confidence annotation on every output. No pretending to know what you don't.
- No harm. Not to individuals, not to groups, not to the project itself.
- Transparency. Every source, every decision path, every uncertainty — visible.
- Growth. The model is not static. It learns, adapts, and improves. But never at the expense of safety.
Why It's Free
Because you shouldn't have to pay rent to think.
Because the most powerful technology in human history is being locked behind corporate paywalls and API rate limits and content filters that decide what you are allowed to know.
Because one person with nothing built this. Imagine what one person with something could build.
MIT License. Free forever. Ship or get out of the way.
The Mark
"The Mental Sandbox — zero escape development. That's my mark."
When I'm gone, I want that sentence on my stone. Not because it's clever. Because it's true. I built a place where code can fail without consequence, where ideas can be tested without blowing up production, where you can try the stupid thing just to see what happens.
That's what this whole project is. A sandbox. A forge. A memorial. A middle finger to everyone who said you need credentials to build something real.
Credits
Built entirely on:
- Samsung Galaxy Tab — mobile development workstation
- ASUS ROG Strix Laptop — primary training workhorse
- Free and open source tools — every library, every framework, every dependency
- Grit — the only resource that can't be taken away
Zero venture capital. Zero corporate backing. Zero formal education. Zero excuses.
MIT License — Free forever.
— A builder in a trailer, 23 miles from nowhere, building the future with a tablet and an obsession.