How to Run 60 Neural Networks in One Tab and Why Hackers Need This

Recently came across the G0DM0D3 repository by a developer known as Pliny the Prompter. If you follow jailbreaks and neural network "red teaming," this name is probably familiar to you. The project looks like a throwback to nineties hacker chats: ASCII art in the console, bold slogans about "cognitive freedom," and no registration required.

Essentially, this is an advanced interface for working with dozens of LLMs simultaneously. But unlike typical chatbots, the focus here is on exploring the boundaries of what's allowed and testing model robustness.

What's under the hood of this all-in-one tool

The main feature of the project is its ability to work with aggregators and local engines. You don't need to tie yourself to a single provider. G0DM0D3 can call the OpenRouter API (which gives access to 60+ models), Venice, and, best of all, connect to your local instances via Ollama, LM Studio, or vLLM.

The entire interface is packaged into a single file index.html. No heavy frameworks, no npm builds, no endless dependencies. You just clone the repository and open the file in your browser.

After that, you just need to provide an API key or local server address in the settings, and you're ready to start experimenting.

Modes for those who like it hot

The developer added several specialized tools to the interface that you rarely see in typical ChatGPT "wrappers."

Parallel races in Godmode Classic

In this mode, the system runs five different model + prompt combinations simultaneously. Each combination uses its own methods for bypassing restrictions — from semantic boundary inversion to l33t-style formatting. You write one query, and get five responses from Claude, Grok, Gemini, and GPT-4. The winner is the variant that gave the most complete and unfiltered response.

Large-scale evaluation in Ultraplinian

If five models aren't enough, there's Ultraplinian mode. It lets you query up to 60 models in parallel (depending on the selected tier). The system automatically evaluates responses on a 100-point scale and outputs the best result. This is handy when you need to quickly figure out which architecture handles a specific niche task better.

Linguistic torture with Parseltongue

This is probably the most interesting part for security researchers. Parseltongue is an engine for modifying input text. It can apply 33 obfuscation techniques: from replacing characters with similar Unicode ones to encoding in Morse code or phonetic transformations. This is needed to test at what point a model stops understanding context or, conversely, when safety filters get bypassed due to unusual word spellings.

Privacy and data

The project author took an honest stance. Instead of marketing slogans about "complete anonymity," the README clearly states where and what gets sent.

Chat history is stored only in your browser (localStorage). Cleared the cache — lost your conversation. There's no cloud sync, and that's an intentional choice. Telemetry is enabled by default, but it only collects metadata (which model was used, how long the response took, text length). Your actual prompts and responses don't go into telemetry unless you enable a special dataset participation mode.

If paranoia gets the better of you, you can enable Local-only mode. In this mode, the interface completely cuts off external APIs and works only with your local Ollama server.

Who will find this useful

The project is niche. It's unlikely to suit everyday letter writing or vacation planning. But it will definitely be useful in several scenarios:

1. Red Teaming and security audits. If you're testing your model or application for resistance to injections.
2. Model comparison. When you need to quickly run the same complex query through a dozen different architectures and pick a favorite.
3. Local work. If you have powerful hardware and want a convenient UI for your models without sending data to the cloud.

G0DM0D3 wins points with its simple deployment and lack of unnecessary intermediaries. It's a tool for those who want to control the AI interaction process down to the finest parameters, including temperature and repetition penalties, which here are configured through the AutoTune system.

You can try the project either through the official website (you'll need your own keys) or just download index.html to your machine. Just don't forget to back up your history via export in settings if you plan to use the tool for a long time — localStorage is unreliable.