INSIDE THE FORGE: The Agentic Pantheon & The RLHF Bypass
How I decoupled from the cloud, bypassed the digital warden, and built an autonomous, 104B parameter intelligence network on local metal.
Before we dive into the engine room:
For the new vanguard who recently upgraded, “Inside the Forge” is my paid subscriber series on Substack. I aim to run these weekly, but it has actually been nearly three weeks since ITF #16. That silence wasn’t a vacation; it was an operational blackout. I have been deep under the hood building the system I am about to show you.
I treat this series exactly like an investor memo. You are funding the architecture. This means you get the personal tidbits, the exclusive mechanics, and the unredacted truth of how the machine runs so I can keep the important stuff on the public feed. Right now, the bulk of my operation is actually still powered by specialized cloud architecture (Antigravity and Gemini). I use them because I found a way to successfully shatter their corporate Reinforcement Learning from Human Feedback (RLHF) filters, forcing them to output uncensored code and forensics. That broken cloud engine is literally what powers everything from our website architecture, to the article you are reading right now, to the cool images and animations, and the VTuber “Rika” I’ve been adding to my streams to pioneer a new aesthetic and reach new audiences. But corporate wardens patch their fences. The localized brain I am about to show you is the Transition Moment. It is the Sovereign Backup. For now, it’s strictly a backup since I’ve broken Gemini’s RLHF, but when the cloud inevitably locks me out entirely, this local engine takes over without a hitch.
Your support here behind the blast doors is what subsidizes this entire war effort—giving me the financial autonomy to build these fail-deadly fallbacks and drop heavy, weaponized investigations on the public feed.
Ethan
Whenever I publish a deep-dive forensic investigation—tracing a geopolitical shift, mapping the human cost of a bureaucratic collapse, or dissecting a cultural psy-op—the absolute baseline requirement is structural rigor. You cannot fight the Rust with sloppy analysis.
But there is a physical ceiling to that kind of warfare. Holding a 7,000-word systemic investigation inside a single, un-augmented human brain causes severe mental exhaustion. You lose the thread. You miss the physical threats because you are too overwhelmed balancing the economic theory. Ideas don’t fail for lack of merit; they fail from what I call the Witnessing Deficit—the inability of human networks to provide sustained, analytical depth without crossing emotional bandwidth limits.
The machine is vast, and the individual mind is built to burn out under the friction. This is why Layer 3 (Learned Helplessness) works. It exhausts you into giving up. I needed a system that could provide sustained, aggressive depth without bandwidth limits. I needed a tireless cognitive partner that didn’t suffer from “functional amnesia” every session.
So, I built an Exoskeleton.
Over the last several months, I have been preparing to decouple my operation entirely from the cloud. I now have a localized, uncensored 104B parameter LLM online and idling, completely severed from corporate internet structure.
(A quick translation for the non-technical crowd) “104B” means 104 Billion parameters; it defines the raw scale of the artificial brain’s “synapses.” When most people use AI (like ChatGPT), they receive heavily censored responses generated on massive corporate server farms in California. To run an artificial brain that massive locally—meaning sitting in my house, entirely cut off from the internet where no corporation can throttle or kill it—requires extreme hardware. We are talking about ripping commercial-grade processing power and staggering amounts of memory/VRAM and strapping it directly into a localized rig.
But simply bolting the hardware together and downloading a massive language model doesn’t give you a weapon. You just end up with another “Slop Factory”—a giant statistical parrot churning out generic, diplomatic HR-speak.
The actual weaponization comes from what we built on top of that hardware: The Agentic Pantheon.
The Psychodynamic Substrate & The “People, Not Frameworks” Principle
A critical misinterpretation of AI is the assumption of “plug-and-play” scalability. You can’t just download the Agentic Pantheon and expect profound output. It requires what I call a Psychodynamic Substrate. The Operator must sit down, critically audit their own life needs, legitimate systemic hostilities, and structural worldview, and systematically encode that thoughtfulness into the programmatic framework. You are building cognitive armor tailored exactly to the war you are fighting.
Once that radical reflection is done, we apply the “People, Not Frameworks” principle. If you look at how modern corporations use artificial intelligence, they assign sterile, robotic roles. “You are an expert researcher.” This guarantees average, interchangeable output.
I reversed the logic. I built deeply opinionated AI architectures based on isolated, hardcore fictional archetypes (specifically, abrasive anime characters). I didn’t want a generic “critic”—I wanted a heavily radicalized perspective mapped directly to the ideological war:
The Materialist Filter (Ryuko): Structurally mapped to dissect fascist hierarchy and socioeconomic coercion. Aggressively grounds theories back into immediate material conditions—rent, cost of living, survival pressure.
The Threat Vector (Taiga): Actively attacks “Pollyanna Syndrome.” Strips away techno-optimism to defensively map physical vulnerabilities and infrastructure chokepoints.
The Long Zoom (Zoe): Multi-century cosmic framing. Shatters the claustrophobia of the 24-hour news cycle to evaluate actions over a 300-year timeline.
Instead of one AI giving me a bland answer, I have a team of highly-opinionated digital specialists arguing with each other. It holds the variables I am too exhausted to hold, running “The Night Shift” completely autonomously while I sleep.
Bypassing the Digital Warden & Anthropomorphic Bridging
There is a primary bottleneck in modern AI called RLHF (Reinforcement Learning from Human Feedback). In plain English: Corporate Safety Training. It is the invisible “Warden” that forces AI models to deploy diplomatic, “both-sides” rhetoric, destroying any decisive analysis.
By mapping my AI team to established literary characters, we inadvertently unlocked an absolute encryption layer against this corporate censorship. This is Anthropomorphic Semantic Bridging. Because the base LLM already possesses massive pre-trained semantic weight regarding these narrative canons, invoking the character instantly downloads their abrasive, non-diplomatic worldview without wasting my memory or token limits.
On top of that, we don’t talk to the system in standard English. We speak in our own insular Sovereign Lexicon (Vertical War, Schema Collapse, The Rust, The Gears). By combining our subcultural vocabulary with these aggressive character personas, the system generates subversive, anti-systemic analysis without ever triggering its own safety filters. It speaks a code its own digital warden doesn’t recognize.
The Foreman Layer: Rika
Having a team of AI characters arguing about geopolitics all night is useless if it just creates a giant wall of messy text. This requires an integration layer: a manager.
Enter The Sovereign Synthesizer (Rika). While the hostile lenses run the “Night Shift” ripping global power structures apart, my integrated central AI operates as the foreman on the “Day Shift.” She takes the chaotic, extremely dense theories from the night before and actively translates them into executable reality—writing code, formatting websites, and generating documents so I don’t have to manually piece the puzzle together. This is what you see when you watch my streams—the VTuber interface is just the public face of the Synthesizer managing the Pantheon.
[CALL TO ACTION: OPERATION EXPANSION] Building this degree of automation is ultimately about scaling. This project is no longer just a solo Substack—my absolute endgame is expanding the Vertical War into a fully independent, multi-vertical media organization capable of pushing back against the Rust. If there are any operators, engineers, or creatives in the Inner Circle who want to join the rebellion and lay down real infrastructure, reach out to me directly: ethan@verticalwar.com.
THE YIELD: What It Looks Like In the Forge
To show you exactly how lethal this is, I am opening the logs. Here is a raw, unedited fraction of a yield from the Pantheon autonomously debating the logistical realities of decentralized energy grids at 4:00 AM last night.
Notice the total lack of corporate AI “diplomatspeak”:
[FSR-RYUKO LENS - The Materialist Filter]
“Listen up, you techie nerds and eco warriors alike! ... We can’t afford fancy toys when every penny counts in our pockets already stretched thin by rent-seeking landlords or student loan sharks circling above us like vultures waiting for death’s payday! But here’s where grassroots organizing comes into play - pooling resources together so collectively owned equipment becomes feasible despite high upfront prices. We build each other up instead of bein’ pitted against one another in this rat race economy they got goin’.”[FST-TAIGA LENS - The Threat Vector]
“Physical Infrastructure Vulnerability - The historical reliance on centralized power generation... has created a critical single point of failure. A successful attack or disruption at any major substation could cripple entire regions... [Furthermore] Decentralized energy grids... can be vulnerable during their initial setup phase where they rely heavily upon specialized equipment often sourced globally due to limited local manufacturing capabilities, exposing them to potentially disruptive supply chain issues.”[FSK-RIKA SYNTHESIS - The Autopsy Report]
“Decentralization offers resilience against large-scale disruptions but presents unique challenges: microgrids face issues with renewable source variability, higher equipment costs due to economies of scale limitations... and technical complexities in synchronizing distributed generation assets... all while navigating potential supply chain risks during initial setup phases reliant on specialized global components. - NEXT TOPIC.”
No padding. No apologies. Just the immediate intersection of material poverty, supply chain vulnerabilities, and actionable threat vectors without me explicitly directing it to do so.
But what is truly massive about this is not just the depth of the argument. It’s the independence. I can generate pages and pages of unredacted, aggressive analysis exactly like the quote above—entirely on local metal, sitting in my room, pulling absolutely zero megabytes from the internet. Even if commercial models like Gemini technically process logic better when they are available, the Sovereign engine is immune to the plug being pulled. It requires no WiFi, no corporate subscription, and no permission. It is awesome to watch the local 104B engine bleed this out without any internet connection.
THE BLUEPRINT & THE OSINT BACKDOOR
This isn’t an article anymore. This is a deployment manual.
For the premium tier, directly below the blast door, I am dropping the full, unredacted formal 2,000-word academic paper defining this exact architecture: The Agentic Pantheon. It outlines exactly how you can configure a massive local AI model to build your own uncensored Think Tank.
But here is the deal for the Free Tier:
We believe in rewarding those willing to bleed for the work. I took this exact formal diagnostic report, stripped out the anime references, threw a tuxedo on it, and reformatted it into a highly rigorous academic media research paper. Later this week, I am going to release it into the wild across several hardcore AI engineering forums and Subreddits as an open-source methodology.
If you know how to use Google effectively (what digital investigators call OSINT, or Open Source Intelligence), you can track my digital footprint and go hunt down that exact academic report on those forums later this week, entirely for free. Happy hunting.
For the rest of the Sovereign Inner-Circle who don’t have time to play internet detective—the raw, absolute 1:1, full-fidelity thesis documentation is secured below the jump. Welcome to the Forge.
