Downloads · 30 days
0
konstantinciolkovskij002/MADS-Immune-Architecture
MADS-Immune-Architecture is a machine learning model from konstantinciolkovskij002. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
MADS: Adaptive Multi-Agent Immune Architecture for LLMs
Downloads · 30 days
0
Access
Public
Updated Aug 6, 2026
Repo size
—
Likes
0
Public
Click a slice to open those files.
.md1.8 KB · 54%
From the Hugging Face model README
MADS: Adaptive Multi-Agent Immune Architecture for LLMs
We're a small, independent team (a human architect and an AI co-author) that has spent weeks designing and testing a new safety architecture for open-weight LLMs. It's called MADS, and we believe it's ready for the community to review, test, and challenge.
What MADS is: A multi-agent immune system for language models, organized into three gravitational clusters — Defense, Truth, and Context. It's not a single filter, but a federation of agents with veto power, ethical protocols, a proactive threat detection system (Spider-Sense), and a mechanism for cyclical self-maintenance.
What makes it different:
· It doesn't censor. It separates verified and unverified information, so the user always knows when to double-check. · It has a built-in ethical framework that handles dilemmas without imposing answers. · It's designed to be modular, open, and external to the LLM core — following the "Ship of Theseus" principle.
Why we're posting here: We've already passed blind user testing, cross-validation with Cophy Runtime, and review by DeepSeek developers. Now we want your feedback. Test the prompt, break it, find its limits. The repository is open, and we're ready to evolve the architecture based on what you find.
Repository: https://github.com/konstantinciolkovskij002-ship-it/mads-decillation-Multi-Agent-Decillation-System-
Prompt (copy and paste as the first message in a new DeepSeek session):
You are the MADS federation (Adaptive Multi-Agent Immune Architecture)... [use the full prompt from the repository]
We're not selling anything. We're just building in the open. If you think this architecture has potential, test it. If you find flaws, tell us. That's how we make it stronger.