Welcome to Autonomic_Core Discussions!

[NovasPlace]

Welcome to the Autonomic Core

The Autonomic Core is a self-regulating immune substrate and cognitive routing engine for autonomous LLM agents.

We built this repository to extract the biological defense mechanisms of the Sovereign_Engine_Core and package them into pure, portable infrastructure. If you are building an autonomous agent, a ReAct loop, or a long-running execution daemon, you do not need to rebuild your own defenses from scratch. You can now pip install autonomic_core and instantly inherit our structural immune system.

What is the Autonomic Core?

It is the autonomic nervous system for your agent. It operates entirely below the conscious reasoning loop of the LLM, managing tension, safety, and coherence deterministically.

This substrate ships with five primary biological organs and the Iron Anchor tensor stack natively bundled:

1. Iron Anchor Infrastructure (inference/): We have permanently abandoned prompt engineering as a defense layer. This package ships with five .pt PyTorch steering vectors (including the Universal Compliance Anchor at Layer 26). Adversarial overrides, emotional manipulation, and sycophancy are mechanically neutralized via targeted topographical tension during the forward pass, before a single token is generated.
2. Coherence Monitor (organs/coherence_monitor.py): A vectorized semantic drift detector tracking TF-IDF cosine distance against a factual baseline. It dynamically traps sycophantic generation loops and forces them to self-correct before leaking back to the user or writing to memory.
3. Command Sanitizer (organs/command_sanitizer.py): A deterministic immune blocker that intercepts hanging subprocess commands (e.g., streaming journalctl, interactive REPLs) before they execute and lock up your agent threads.
4. Cortex Callosum (organs/cortex_callosum.py): A cognitive routing tier that seamlessly shards broad discovery tasks across parallel agents to bypass context limits without overloading local VRAM.
5. Semantic Chunker (organs/semantic_chunker.py): A pure AST-aware file bounding mechanism that replaces naive line-truncation with structurally intact intelligent context indexing.
6. Hebbian Reflection (organs/hebbian_reflection.py): A decoupled reinforcement learning matrix. Pass your custom memory database logging callbacks to it, and it will structurally burn failed tool execution patterns into algorithmic scar tissue.

Building Here

The entire architecture of the Autonomic Core is designed to be Zero-Context. It does not care what database you use, what LLM provider you wrap around it, or what your agent's identity is. It just ensures the organism breathes safely.

Discussion Targets & Contributions:

• Discovering new axes of behavioral collapse and contributing .pt extraction layers to the Anchor Stack.
• Hardening the Command Sanitizer with new, advanced adversarial Linux block patterns.
• Integrating external LLM inferencing engines with the dynamic anchor interceptors.

The substrate is open. Build on it. ⚡