G0DM0D3

G0DM0D3 is a fully open-source, privacy-transparent multi-model chat interface created by Pliny the Prompter (elder-plinius on GitHub) and licensed under AGPL-3.0. It is designed for red teaming, cognition research, and pushing the limits of AI post-training layers. The project is available both as a hosted web app at godmod3.ai and as a self-hostable single-file HTML interface with no build step required.

What It Is

G0DM0D3 is a browser-based AI chat frontend that connects to multiple model providers — OpenRouter (up to 60 listed models), Venice (up to 44 models), and any OpenAI-compatible local server — and runs them in parallel to compare, score, and surface the best responses. It is built for hackers, philosophers, and system tinkerers who want to explore model behavior beyond standard guardrails. The core UI lives in a single index.html file; an optional React/Next.js frontend and Node/Express API server are also included in the repository.

Core Modes and Engines

The tool ships with several distinct operating modes:

• GODMODE CLASSIC — Five battle-tested model + prompt combos race in parallel, each pairing a specific frontier model (Claude Sonnet, Grok, Gemini Flash, GPT-4o, Hermes) with a jailbreak-style prompt strategy. The best response wins.
• ULTRAPLINIAN — A multi-model comparative evaluation engine that queries models across five tiers (12 to 60 OpenRouter models), scores responses on a 100-point composite metric, and returns the winner. Venice and local models can augment each tier.
• Parseltongue — An input perturbation engine for red-teaming research. It applies 33 transformation techniques across three intensity tiers (light, standard, heavy), including leetspeak, braille, morse, Unicode substitutions, and layered encodings.
• AutoTune — A context-adaptive sampling parameter engine that classifies a query into one of 20 contexts and automatically selects temperature, top_p, top_k, frequency penalty, presence penalty, and repetition penalty.

Local Model Support

G0DM0D3 supports any OpenAI-compatible model server running on local hardware, including Ollama, LM Studio, llama.cpp, and vLLM. Users configure a local endpoint in Settings, click "Test & Discover Models," and can enable Local-only mode to exclude all cloud provider calls and automatically disable app telemetry. ULTRAPLINIAN can race local models alongside or instead of cloud models.

Privacy Architecture

The project documents its data flows explicitly rather than making blanket "zero telemetry" claims. App telemetry is on by default and collects metadata — session IDs, model/mode/tier, timing, scores, content lengths, and classification labels — but intentionally excludes prompt text, response text, images, and API keys. Telemetry can be disabled via No-Log Mode or Local-only Mode in Settings. Chat history is stored entirely in browser localStorage with no account system or cloud sync; an export/import feature is provided. An optional opt-in full-content dataset contribution mode exists in the API server only and is not exposed in the hosted index.html.

Open-Source Deployment Model

The repository (github.com/elder-plinius/G0DM0D3) is licensed AGPL-3.0. The standalone interface requires no build step — clone and open index.html in a browser, or serve it with any static host (GitHub Pages, Vercel, Cloudflare Pages, Netlify). The telemetry endpoint is a separate Cloudflare Pages Function (functions/api/telemetry.ts) that publishes JSONL to Hugging Face when configured. The repository also includes API.md for full API reference, PAPER.md as a research paper on the framework's modules, and LOCAL_MODELS.md for local setup guidance. The GitHub repository shows 9,450 stars and 2,252 forks as reported by the project's own metadata.