Skip to content

Systems Orchestration (multi-paradigm architecture)

An architecture for routing and combining LLM nodes using Physics, Biology, Economics, and Music.

What is this?

Rather than viewing LLM agents as isolated chat interfaces or sequential tool-chain links, this module orchestrates them as components of a dynamic, multi-paradigm system. Based directly on the "Systems & Intelligence" Architect Prompt, the architecture mathematically routes compute and handles consensus via four paradigms:

  1. Harmonic (Music): Agents are oscillators. The system calculates the pairwise cosine similarity of their hidden utility vectors (Interaction Matrix \(\mathbf{M}\)) and runs eigenvalue analysis to find the "dominant melody." Used for consensus and brainstorming.
  2. Homeostatic (Biology): Agents are cells. The system audits their Von Neumann-Morgenstern (VNM) Transitivity. If Coherence \(C\) drops, restorative feedback enforces stability.
  3. Market (Economy): Agents are bidders. Compute resources are allocated based on marginal utility per task.
  4. Flow (Physics): Agent communication is a gradient field seeking the path of least entropy.

Module Structure

  • core/utility.py: The mathematics. Calculates VNM Coherence (\(C\)) from preference graphs, derives the Utility Vector \(U\), and computes Harmonic Resonance matrices.
  • orchestration/conductor.py: The routing mechanism. Contains the ParadigmSwitcher meta-controller which takes a task_context and dynamically routes execution to the Harmonic, Homeostatic, Market, or Flow implementations.
  • agents/manager.py: The LLM interface. Translates raw textual preferences out of an LLM into the structural variables (\(U\) and \(C\)) used by the orchestrator.

Usage

To see the Multi-Paradigm architecture in action, run the live demonstration script. This will spin up three virtual LLM agents (with diverging utility functions) and route tasks dynamically between the Harmonic, Homeostatic, Market, and Flow paradigms:

python3 live_demo.py