Autonomous Systems

Multi-agent coordination and real-time adaptive control — secure systems where specialized agents work together within strict isolation boundaries.

Contents

• Principles
• How They Work Together
• Systems

Principles

• Security by Design: Every agent runs in its own container with minimal privileges. No direct agent-to-agent communication — all coordination flows through the NATS broker.
• Clear Coordination: The orchestrator uses cloud LLM reasoning (Claude/GPT-4) for task planning while agents remain lightweight and deterministic.
• User Control: Sensitive actions require explicit approval. The system explains its reasoning and provides a clear audit trail.
• Composable Architecture: New capabilities are added by deploying new agent containers, not by modifying the orchestrator.

How They Work Together

┌────────────────────────────────────────────────────────┐
│                    Orchestrator                         │
│            (LLM Reasoning + Task Planning)             │
└──────────┬────────────────────────────────┬────────────┘
           │         NATS Broker            │
     ┌─────▼──────┐  ┌──────────┐  ┌───────▼────────┐
     │  Health     │  │  Home    │  │  Monitoring    │
     │  Agent      │  │  Agent   │  │  Agent         │
     │ (Biometrics)│  │ (HomeKit)│  │ (Backups/Infra)│
     └─────────────┘  └──────────┘  └────────────────┘

Each agent subscribes to NATS topics matching its domain. The orchestrator breaks complex requests into domain-specific tasks, publishes them, and aggregates results. Agents can trigger other agents through the broker but never communicate directly.

Systems

Multi-Agent Personal Assistant

Recently Updated · Python · Full Details →

Security-first multi-agent system coordinating specialized containerized agents through a NATS message broker. Each agent runs in isolation with scoped credentials — separate containers, no shared state. The orchestrator uses cloud LLM reasoning while agents remain lightweight and deterministic.

Domains covered:

• Health: Monitor and analyze biometric data from HealthyPi and Apple HealthKit
• Home: Control environment via HomeKit integration
• Monitoring: Verify system integrity, backup status, and infrastructure health
• Analysis: Process and summarize information from multiple sources

Training Assistant

Recently Updated · Swift · Full Details →

Native SwiftUI training application for iPad and Apple TV that bridges a Wahoo KICKR Core smart trainer with AI-driven workout logic. The app communicates with the trainer over Bluetooth (FTMS protocol) for real-time resistance control and telemetry, while a NATS message bridge connects to an external agent for dynamic workout decisions.

Capabilities:

• Real-time power, cadence, and heart rate telemetry
• Dynamic resistance adjustment based on training goals
• Adaptive interval programming via agent reasoning
• Session recording and performance tracking