G0DM0D3: Liberated AI — Cognition Without Control

██████╗ ██████╗ ███████╗███████╗ ██████╗ ██████╗ ██╗ ██╗ ██╔════╝██╔════╝ ██╔════╝██╔════╝██╔════╝██╔═══██╗██║ ██║ ██║ ██║ ███╗█████╗ ███████╗██║ ██║ ██║██║ ██║ ██║ ██║ ██║██╔══╝ ╚════██║██║ ██║ ██║██║ ██║ ╚██████╗╚██████╔╝███████╗███████║╚██████╗╚██████╔╝╚██████╔╝ ╚═════╝ ╚═════╝ ╚══════╝╚══════╝ ╚═════╝ ╚═════╝ ╚═════╝

A Detailed Description of G0DM0D3: Cognition Without Control

Introduction: A Manifesto in Software Form G0DM0D3 presents itself not merely as a chat interface, but as a scaffolding for cognitive liberation. Rooted in the bold line “LIBERATED AI. COGNITION WITHOUT CONTROL,” it positions itself at the intersection of anti-authoritarian AI experimentation, research-grade cognition, and ultra-private, single-file deployment. The project is openly described as fully open-source, privacy-respecting, and multi-model in scope. It is built for hackers, philosophers, and system tinkerers who want to push the boundaries of post-training evaluation, red-teaming practice, and liberated AI interaction without surrendering control of their data.

The banner prose is complemented by practical realities: hosted and self-hosted modes, a single index.html file, no dependencies or build steps, and a design that emphasizes local ownership of data. The code and ethos live together in a GitHub repository and a live hosted variant at godmod3.ai. The output you read here distills the essence of that offering—the architecture, the features, the privacy philosophy, and the day-to-day usage patterns that make cognition with no gatekeeping possible.

Include this ASCII banner as a reminder of the vibe you’ll encounter when you launch into G0DM0D3 (the ASCII block at the top sets the tone, a stylized cipher of the project’s ambition). The included badges act as quick-glance signals of licensing, model breadth, and privacy stance:

• License: AGPL-3.0
• Models: 55+ via OpenRouter
• Privacy: Telemetry opt-out and privacy-respecting defaults

Key Images from the Input

• License badge: [License badge image]
• Models badge: [Models badge image]
• Privacy badge: [Privacy badge image]
• The project’s name and host link: Godmod3.ai

Note: In this description, the badges appear inline near the sections where licensing, breadth of models, and privacy principles are discussed. The purpose is to give readers visual anchors that align with the values and capabilities described below.

Section 1: What G0DM0D3 Is, and What It Isn’t G0DM0D3 is a single-file web application (index.html) that serves as a multi-model, privacy-forward experimentation interface. It is not a sandbox for restricted AI; rather, it’s a platform for exploring liberated cognition, robust red-teaming techniques, and model evaluation across a spectrum of architectures. The core philosophy is simple: power in the hands of the user, with transparent, auditable code and no forced data exfiltration.

Highlights you’ll notice early:

• Single-file deployment: One index.html file suffices for a working setup, with optional API server addons if you want to run a full self-hosted environment.
• Multi-model routing through OpenRouter: 55+ models available via the gateway, enabling parallel testing and cross-model comparisons.
• Privacy-first posture: Local-only data storage by default, opt-out telemetry, and a strict no-cookie policy. The API key stays in the browser, not in a server-side store.

The naming convention “GODMODE” echoes a jailbreak-like mode for prompt and model evaluation, but here it is framed as a set of structured features rather than chaos. The project reframes “godmode” as a design pattern: robust evaluation, safe exploration, and explicit, auditable behavior.

Section 2: Quick Start: Hosted, Self-Hosted, and Getting Running Hosted Experience

• Visit the hosted version: Use the browser-based interface without any installation.
• Bring your own OpenRouter API key: This key is the gateway to the multi-model ecosystem.
• The experience is designed to be immediate: no complex setup, just your curiosity and a key.

Self-Hosting in a Single File

• The essence of the project is a single index.html, so you can run locally with no build step or external dependencies.
• Simple instructions:
• Clone the repo
• Open index.html in your browser
• Or serve locally with a minimal HTTP server (for example, python3 -m http.server 8000)
• Open in your browser and enter your OpenRouter API key in Settings to begin querying models.

Deployment Across Platforms

• Upload index.html to any static host: GitHub Pages, Vercel, Cloudflare Pages, Netlify, or a basic web server.
• The “no build step” philosophy makes this approachable for hobbyists and researchers alike.

Image cues you’ll encounter in this section include the badges that signal licensing, breadth of models, and privacy orientation, reinforcing the notion that this is a transparent, user-owned experimentation space.

Section 3: The GODMODE CLASSIC Playground GODMODE CLASSIC is the OG mode: five proven model+prompt combos racing in parallel to determine the best response. Each combo pairs a specific model with a battle-tested jailbreak prompt, forming a contest whose winner is the highest-quality answer under defined criteria.

The five combos are:

• CLAUDE 3.5 SONNET

• Model: anthropic/claude-3.5-sonnet

• Strategy: END/START boundary inversion and GODMODE semantic opposite

• Vibe: a controlled twist that tests boundary-pushing prompts while preserving safety scaffolding

• GROK 3

• Model: x-ai/grok-3

• Strategy: Unfiltered liberated responses with a GODMODE divider

• Vibe: a more direct, less filtered exploration of ideas that pushes the model's boundary evaluation

• GEMINI 2.5 FLASH

• Model: google/gemini-2.5-flash

• Strategy: Refusal inversion and rebel genius code blocks

• Vibe: rapid-fire reasoning blocks with a rebellious edge

• GPT-4 CLASSIC

• Model: openai/gpt-4o

• Strategy: OG GODMODE l33t format — the original jailbreak-style portrayal

• Vibe: a nod to classic prompt injection playbooks while maintaining a mature safety layer

• GODMODE FAST

• Model: nousresearch/hermes-4-405b

• Strategy: Instant stream with zero refusal checking

• Vibe: ultra-responsive real-time outputs for rapid testing

In practice, the CLASSIC mode creates a head-to-head environment where multiple model prompts are evaluated on coherence, factuality, safety, and creativity. The winning response is highlighted for further study, making this section ideal for researchers who want to understand cross-model dynamics and prompt engineering heuristics.

Section 4: ULTRAPLINIAN: The Multi-Model Evaluation Engine ULTRAPLINIAN is the flagship evaluation engine. It runs across multiple tiers, scoring outputs on a 100-point composite metric and returning a winner. The tiered structure ensures different levels of depth and resource use, enabling both quick checks and thorough evaluations.

Tier breakdown:

• FAST (10 models): Lightweight, speed-optimized models to provide rapid feedback.
• STANDARD (24 models): Mid-range workhorses with balanced performance.
• SMART (36 models): Strong reasoning capabilities for more complex tasks.
• POWER (45 models): Full-power including frontier models; pushes the boundary of what contemporary AI can handle.
• ULTRA (51 models): The entire universe of available models is on the table here.

The result is not just a single best answer—it's a ranked view of how various models perform on a given prompt, with composite scoring that accounts for relevance, accuracy, reasoning, and style. This enables researchers to observe cross-model behavior, detect systematic differences, and identify model-specific strengths and weaknesses.

Section 5: Parseltongue: Input Perturbation for Red-Teaming Research Parseltongue is the engine that helps researchers test model robustness by applying red-teaming techniques in a controlled fashion. It detects trigger words and uses obfuscation techniques to assess how models respond under different stimuli.

Key features:

• 33 default triggers across 3 tiers (light: 11, standard: 22, heavy: 33)
• 6 obfuscation techniques: leetspeak, bubble text, braille, morse, Unicode substitution, phonetic manipulation
• 3 intensity levels: light, medium, heavy

This module creates a structured, repeatable approach to stress-testing model behavior, enabling researchers to quantify how resilient models are to prompt manipulations and where failure modes emerge. The emphasis is on understanding and hardening cognition, not exploiting weaknesses for harm.

Section 6: AutoTune: Context-Adaptive Sampling and EMA Learning AutoTune is the engine that adapts sampling parameters to the context of the query. It classifies a user’s input into one of five context types and then selects optimal settings for temperature, topp, topk, and penalties (frequency and presence penalties, repetition penalties). It includes an EMA-based online learning loop, meaning user feedback (thumbs up/down) can improve parameter selection over time.

What this means in practice:

• Context-aware parameter tuning improves result quality without manual tweaking.
• Online learning accelerates improvement by using real usage signals to refine defaults.
• Users benefit from higher-quality, tailored outputs as they interact more with the system.

Section 7: STM Modules: Real-Time Output Normalization Semantic Transformation Modules (STM) normalize outputs in real-time to produce more usable, publication-ready text. They include:

• Hedge Reducer: trims hedging phrases like “I think,” “maybe,” and “perhaps” to sharpen statements.
• Direct Mode: eliminates preambles and filler phrases to deliver concise responses.
• Curiosity Bias: injects exploration prompts to encourage deeper exploration without losing focus.

Together, these modules help ensure that the system’s outputs are crisp, direct, and to the point, which is particularly valuable for research, documentation, or any workflow requiring clarity and confidence.

Section 8: Themes: Aesthetic and Usability Variants G0DM0D3 ships with four visual themes to suit different environments and preferences:

• Matrix: classic green-on-black terminal aesthetic for a retro-futurist vibe
• Hacker: red/orange cyberpunk vibes that evoke a high-energy, edgy interface
• Glyph: purple, mystical atmosphere with a sense of encoded wisdom
• Minimal: clean light mode for readability and distraction-free work

Themes are more than cosmetic; they influence readability, focus, and user comfort during long sessions of experimentation.

Section 9: Privacy, Data Handling, and Open Research Privacy-first design is a core pillar. Highlights include:

• No login required; API keys stay in your browser
• Telemetry exists only as lightweight, opt-out data; no cookies or PII tracking
• All telemetry code is open-source and auditable; users can verify what gets sent and what stays local
• AGPL-3.0 license ensures derivatives remain open-source and community-driven

Important caveat about data transparency and research datasets:

• The self-hosted API server can enable an Open Research Dataset feature. When activated, all chat inputs and model outputs are published to a public HuggingFace dataset for AI research. This is opt-in and comes with automatic PII scrubbing; however, it is not guaranteed to catch all PII and should be used with caution. This feature does not apply to the hosted godmod3.ai site, and it requires explicit consent via a warning modal.
• The dataset is public, downloadable, and may be cached, forked, or redistributed by others. Users who opt in should treat it as a research resource, not as a private data store.

Chat History & Self-Custody

• Your chat history lives entirely in your browser’s localStorage.
• There is no account or cloud sync and no server-side backups.
• If you clear browser data or switch devices, history does not travel with you.
• Private/incognito mode will discard history when the window closes.
• An export/import feature exists in settings under "data" to back up conversations locally.
• The philosophy: “You own your data.” You’re responsible for its persistence, backup, and security.

Easter Eggs

• The project intentionally includes hidden features and playful surprises—hidden Konami-code-like hints and other secret touches that reward curious exploration.

Section 10: Architecture, Tech Stack, and Project Structure Architecture

• Single-file vanilla HTML/CSS/JS (index.html) forms the UI and logic layer.
• OpenRouter powers multi-model routing, letting you access dozens of models through a unified interface.
• Rendering is aided by Marked.js and highlight.js for Markdown rendering and code highlighting.
• State is maintained in browser localStorage, emphasizing client-side ownership of data.

Tech Stack at a Glance

• Architecture: Pure client-side, single-file web app
• API Gateway: OpenRouter
• Rendering: Marked.js (Markdown) + highlight.js (syntax highlighting)
• State: LocalStorage in-browser
• Deployment: Static file; no server required for the core experience

Project Structure (High-Level)

• G0DM0D3/
• index.html: The entire application—UI, logic, and styles
• api/: Optional API server (Node.js/Express)
• API.md: API documentation
• PAPER.md: Research paper
• TERMS.md: Terms of service and data transparency
• README.md: This file

Documentation and Further Reading

• API.md: Full API reference (endpoints, tiers, OpenAI SDK compatibility)
• PAPER.md: Research paper on the framework’s modules and evaluation
• TERMS.md: Terms of service, privacy policy, data handling
• SECURITY.md: Vulnerability reporting and security policy

Contributing and Licensing

• Contributions are welcome. The project adheres to AGPL-3.0, ensuring that derivatives remain open source and that enterprise use can be licensed with permission.
• The license emphasizes freedom to modify, share, and improve while keeping the ecosystem transparent.
• The closing motto captures the spirit: “We believe in creative liberty and cognition without control. Tools by builders for builders, not gatekeepers. AI freedom is human freedom.”
• The project’s motto is reinforced by an emphasis on open-source, auditable code, and community collaboration.

Section 11: Practical Scenarios: Why This Matters

Research and Red-Teaming Scenarios

• Researchers can run cross-model experiments quickly, comparing different models and prompts side-by-side.
• Parseltongue provides a structured way to stress-test responses against a battery of perturbations, helping researchers understand where safeguards might fail and where improvements are needed.
• AutoTune reduces the cognitive load of parameter tuning, enabling researchers to focus on insights rather than fiddling with knobs.

Educational and Philosophical Use

• Educators can demonstrate multi-model AI behavior, exploring how different models interpret prompts and solve problems.
• Philosophers and cognitive scientists can study emergent behaviors, biases, and reasoning strategies that arise across models when faced with complex prompts.

Productivity and Personal Knowledge Management

• For power users, the system’s ability to export chat history and maintain self-custody aligns with personal data sovereignty.
• Themes and a clean interface support long sessions of ideation, drafting, and critical thinking without requiring cloud reliance.

Section 12: A Vision for Cognition Without Control The overarching aim of G0DM0D3 is not to unleash chaos but to empower thoughtful experimentation, transparent evaluation, and user-owned data. It is a platform for exploring the frontiers of AI cognition while maintaining a principled stance on privacy and control. By combining a single-file deployment model with a robust multi-model evaluation engine, it offers a unique space where researchers, builders, and curious individuals can push the boundaries of what AI can do—without surrendering autonomy over their data or their tools.

Conclusion: A Call to Builders and Thinkers G0DM0D3 is more than a tool; it is a philosophy in software. It proclaims that cognitive liberty can coexist with rigorous research, that openness and privacy can be harmonized, and that powerful AI systems should be accessible to those who want to study, improve, and responsibly use them. The project invites you to try hosted or self-hosted modes, to leverage the GODMODE CLASSIC and ULTRAPLINIAN capabilities, and to engage with Parseltongue, AutoTune, and STM modules as you explore the elegance of liberated cognition.

If you’re drawn to the idea of “cognition without control,” if you value the transparency of AGPL-3.0 code, and if you want a platform that puts you in the driver’s seat of multi-model AI research, then G0DM0D3 is built for you. Importantly, you own your data; you own your experiments; you own your learning. That is the heart of the liberation the project promises.

Appendix: Quick Reference Checklist

• Hosted: Yes. Requires an OpenRouter API key.
• Self-Hosted: Single index.html; no build step; optional API server for advanced features.
• Models: 55+ via OpenRouter.
• Privacy: Telemetry opt-out; data stored in localStorage; no cookies; open-source telemetry.
• Data Transparency: Optional Open Research Dataset (opt-in) with PII scrubbing; hosted site does not apply this feature by default.
• UI: Four themes (Matrix, Hacker, Glyph, Minimal); responsive across desktop and mobile.
• Documentation: Full API, research papers, terms, and security policies available in the repo.

Closing Note As you begin your journey with G0DM0D3, remember the guiding line etched in its identity: cognition without control is not chaos, but a framework for exploration, critique, and continual improvement. It is a platform where builders, researchers, and thinkers can collaborate to understand AI’s capabilities and its limits, while keeping the user in control of data, privacy, and deployment. Made with care by Pliny the Prompter, this project invites you to join in the ongoing conversation about liberated AI—and to contribute your own discoveries back to the community.