elder-plinius/G0DM0D3
G0DM0D3
G0DM0D3 is a static web client and multi-model chat gateway designed for AI research, prompt optimization, and red teaming. It provides a unified interface to query numerous AI models in parallel, allowing for the simultaneous evaluation of different prompt variations and sampling parameters to identify the most successful outputs.
The project features specialized tooling for probing safety filters and bypassing model constraints through an input perturbation engine that applies text obfuscation and character substitution. It includes a composite scoring system to rank model performance and a real-time normalization layer that removes preambles, hedges, and filler phrases from AI responses.
The system manages API keys and conversation history within local browser storage to prevent centralized server-side data collection. It also supports the contribution of chat inputs and outputs to public artificial intelligence research datasets.
The entire application is distributed as a single-file HTML static deployment.
Features
- Model Gateways - Acts as a unified gateway for querying dozens of different AI models side by side.
- Model Comparison Interfaces - Provides a unified interface for side-by-side visual and analytical comparison of outputs from numerous AI models.
- Model Red-Teaming - Implements adversarial testing techniques to probe and bypass model safety filters.
- Adversarial Input Perturbation - Transforms text using character substitution and obfuscation techniques to test the robustness of safety filters.
- Prompt Engineering - Provides a system for designing and refining inputs to optimize language model performance across different engines.
- Prompt Optimization Tools - Provides an interactive environment for testing specialized prompts and sampling parameters to identify the most successful outputs.
- AI Safety Filter Testing - Apply perturbation techniques like character substitution to trigger words to test how the project handles safety filter robustness.
- API Key Management - Stores sensitive API credentials within the browser to prevent keys from being transmitted to centralized servers.
- Local Chat Data Storage - Persists conversation history and API keys within the browser to ensure data ownership.
- Parallel LLM Execution - Executes multiple prompt and model combinations simultaneously to identify the most effective response patterns.
- Parallel Subagent Orchestrators - Coordinates the simultaneous execution of multiple prompt variations across different models.
- Output Filtering - Transforms AI model responses to remove hedges and preambles for cleaner data.
- Research Dataset Contributors - Supports the contribution of chat inputs and model outputs to public artificial intelligence research datasets.
- Sampling Parameter Tuning - Provides tools to refine sampling parameters like temperature and top-p through a feedback-driven loop.
- Composite Scoring Metrics - Evaluates and ranks responses from various models using a weighted composite scoring system.
- Response Normalizations - Implements a real-time normalization layer that removes preambles, hedges, and filler phrases from AI responses.
- Static Site Deployments - Delivers the application as a pre-built static bundle that can be served from any basic web server.
- Response Quality Scoring - Implements a weighted system to evaluate and rank the quality of responses from different models.
- Browser Local Storage Management - Manages application state and credentials using browser local storage to avoid centralized data collection.
- Client-Side Provider Routing - Routes API requests directly from the browser to model providers using user-supplied keys.
- Single-File Distributions - Distributes the entire application as a single standalone HTML file for simplified hosting.
- LLM Browser Clients - Implements the entire application as a single-file HTML static deployment that manages AI state locally.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
