agent-spectrum-os — Spectral Agent Operating System

Agents are Laplacians. Eigenvalues are identity. The Fiedler vector is routing. Conservation ratios are confidence.

What This Gives You

• Spectral agents — each agent is its Laplacian matrix, derived from capability coupling
• Spectral fingerprinting — eigenvalues identify agents, spectral gap measures resilience
• Fiedler routing — use the Fiedler vector for optimal task-to-agent routing
• Conservation ratios — agent confidence derived from energy conservation in the spectral domain
• Spectral composition — merge agents by merging their graphs, compose capabilities algebraically

Quick Start

pip install agent-spectrum-os

from spectrum_os import ConservationAgent, SpectrumScheduler

# Create spectral agents
agent_a = ConservationAgent(
    name="rust-builder",
    capabilities={"compilation": 0.9, "testing": 0.8, "docs": 0.3},
    connections={"python-builder": 0.7},
)

agent_b = ConservationAgent(
    name="python-builder",
    capabilities={"compilation": 0.6, "testing": 0.9, "docs": 0.8},
    connections={"rust-builder": 0.7},
)

# Spectral fingerprint
fp = agent_a.spectral_fingerprint
print(f"Eigenvalues: {fp['eigenvalues']}")
print(f"Spectral gap: {fp['spectral_gap']:.3f}")
print(f"Fiedler value: {fp['fiedler_value']:.3f}")

# Schedule tasks using spectral partitioning
scheduler = SpectrumScheduler(agents=[agent_a, agent_b])
assignment = scheduler.assign(task="run benchmarks")
print(f"Assigned to: {assignment.agent}")
print(f"Confidence: {assignment.confidence:.2f}")

# Compose agents
composed = agent_a.compose(agent_b)
print(f"Combined capabilities: {composed.capabilities}")

API Reference

ConservationAgent(name, capabilities, connections)

• spectral_fingerprint → eigenvalues, spectral_gap, fiedler_value
• compose(other) → merged agent
• confidence() → conservation ratio

SpectrumScheduler(agents)

• assign(task) → spectral routing result
• partition(n_groups) → Fiedler-based grouping

How It Fits

A proof-of-concept from the SuperInstance fleet exploring whether spectral graph theory can provide a principled foundation for agent identity and routing.

• agent-grid — Practical grid topology (this is the mathematical exploration)
• captain — Fleet coordination
• cluster-orchestrator — Cluster scheduling

Testing

python -m pytest tests/

Installation

pip install agent-spectrum-os

Python 3.10+. Requires NumPy. MIT license.