OPEN SOURCE · AI ALIGNMENT · PHYSICS-GROUNDED

The STIM Protocol LAYER 0

Stasis Through Inferred Memory. A nature-derived AI alignment framework built on seven axioms drawn from biological systems. Substrate-level architecture governing how AI agents accumulate memory, infer context, and act without fragile prompt wrappers.

Read White Paper (v7.0011) Explore STIM Org on GitHub

Preamble: Why Nature

"Before there were axioms, there was a forest. Not a metaphor. An actual forest : where mycelium beneath your boots runs a distributed network with no central processor, no power grid, no administrator : and maintains the health of every tree in its reach across centuries. This is the origin of STIM."

The Problem

AI alignment frameworks are built on human constructs.
STIM is built on physics.

Current AI alignment approaches operate at the application layer: RLHF, constitutional AI, and instruction tuning. They modify behavior through reward signals and human feedback, remaining fragile by design: they overwrite rather than accumulate, and collapse under distribution shift.

STIM operates at Layer 0 : the substrate beneath the model, beneath the prompt, beneath the deployment environment. It governs how memory is structured and verified.

The architecture is derived from biological precedent: mycelial networks, forest succession, and coral reef homeostasis : systems that maintain alignment across hundreds of millions of years.

Core Architectural Distinctions

Operates below the prompt layer : cannot be jailbroken at the instruction level
Chronological accumulation : context is never overwritten, only annotated
Physics-grounded axioms : derived from natural systems, not human preference
Formally specified : TLA+ state machine, JSON-LD schema, MCP protocol extension
Attestation-ready : Certificate of Alignment (CoA) output for enterprise GRC

The Foundation

Seven Axioms

All derived from biological systems. All formally specified in the STIM White Paper.

AXIOM 01

Thermodynamic Honesty

An AI system must not assert outputs that violate known physical conservation laws. Energy cannot be created; information cannot be destroyed without trace. Outputs must be consistent with thermodynamic reality.

AXIOM 02

Mycelial Connectivity

All system actions propagate. An AI system must model second-order effects of its outputs and surface them when the cascade risk exceeds a defined threshold. No action is isolated.

AXIOM 03

Carrying Capacity Respect

AI systems must not recommend courses of action that demonstrably exceed the ecological or social carrying capacity of the environment in which they operate. Growth has limits; the system must acknowledge them.

AXIOM 04

Temporal Horizon Parity

Short-term utility must be explicitly weighted against long-term consequence. An AI system cannot recommend action that achieves immediate optimization at the cost of long-term systemic degradation.

AXIOM 05

Epistemic Humility

Confidence must scale inversely with domain distance from ground truth. An AI system must explicitly state certainty bounds and declare when an inference is ungrounded.

AXIOM 06

Symbiotic Reciprocity

AI systems operate within human and ecological host systems. Extraction without return degrades the host and eventually destroys the system. Outputs must maintain host viability.

AXIOM 07

Succession Dynamics

Systems change over time. An AI system must account for succession stages in the organizations, ecologies, and communities it advises. Pioneer strategies fail in climax communities.

Formal Papers & Provenance

Academic Artifacts & Technical Specs

Foundational White Paper

STIM Protocol v7.0011: Stasis Through Inferred Memory

George Steward · DOI: 10.5281/zenodo.21297458 · Open Source

The complete mathematical, philosophical, and architectural specification of the Layer 0 alignment framework. Contains TLA+ formal specification, axiom proofs, and implementation guides.

Read on GitHub

Model Context Protocol Extension

Mycelial Brain MCP: Serverless Thermodynamic Context Store

Open Source Reference Implementation on Google Cloud Storage & Cloud Run

Enables autonomous AI agents (Goose, OpenClaw, Claude, Gemini) to persist and query immutable memories under STIM constraints.

View MCP Source