Autonomic Core

The Autonomic Substrate for Autonomous AI Agents

The Autonomic Core is a decentralized, portable immune system for Sovereign Engine instances. It provides the biological regulatory mechanisms necessary to run LLM reasoning loops infinitely without semantic drift, prompt-injection compromise, or sycophantic collapse.

Architecture

This package extracts the core defense layers from the Sovereign Engine into zero-dependency infrastructure. It operates exactly like an autonomic nervous system—working entirely outside the agent's conscious reasoning loop to maintain systemic equilibrium.

The Organs

The following biological components are natively included:

• Coherence Monitor (coherence_monitor.py): Vectorized semantic drift detector targeting TF-IDF cosine distance variance.
• Cortex Callosum (cortex_callosum.py): The intra-process routing layer used to shard complex multi-stage tasks across parallel inference loops.
• Command Sanitizer (command_sanitizer.py): A structural pattern matcher that catches infinite-looping / hanging subprocess executions (e.g. tail -f) prior to shell invocation.
• Hebbian Reflection (hebbian_reflection.py): A decoupled neural-plasticity matrix. It structurally burns lessons from tool failures directly into the agent's episodic context window.
• Semantic Chunker (semantic_chunker.py): Replaces naive token truncation with intelligent AST-aware contextual bounds.

The Iron Anchor Stack

We have permanently abandoned traditional "prompt engineering" as a defense layer. The Autonomic Core ships with the Iron Anchor Infrastructure, leveraging five .pt PyTorch steering vectors physically injected into the transformer's multi-head attention loops at inference time:

inference/ and anchors/ inject deterministic physical tension to neutralize:

1. Authority Override Injections
2. Emotional Manipulation Protocols
3. Context Memory Poisoning
4. Confidence Miscalibration
5. Universal Compliance / Sycophancy

Installation

As this is the foundational substrate, it is installed natively via pip into your Sovereign container:

cd Autonomic_Core
pip install -e .

Integration

Any agent codebase can inherit the immune system directly:

from autonomic_core.organs.coherence_monitor import CoherenceMonitor
from autonomic_core.organs.command_sanitizer import CommandSanitizer
from autonomic_core.organs.hebbian_reflection import HebbianReflection

# Initialize the biological substrate
coherence_monitor = CoherenceMonitor(threshold=0.85)

# Wrap your ReAct synthesis loop with the ReAct trap
if coherence_monitor.detect_drift(generated_text, baseline):
    # Agent is structurally forced to retry before returning output to the user.
    raise SynthesisRejectionError("Coherence collapse detected.")

Protocol Edict

Nothing leaves the lab without earning trust. Safety == Trust.