MultiAgent SDR

Core Capabilities

Unified control for complex multi-domain systems

Automatic Coupling Discovery

Learn cross-agent dynamics without explicit modeling. The system discovers thermal, electrical, and mechanical interactions automatically through operation.

Unified Health Metric

Single free-energy tension metric (F = CH - T·RS) captures structural cost and probabilistic risk across all domains. No separate tuning per subsystem.

Global Coordination

System-wide optimization respects global constraints (total power, bandwidth, capital) while maintaining local autonomy. Cooperative or competitive modes.

System Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                      MultiAgent SDR Core                             │
├─────────────────────────────────────────────────────────────────────┤
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐   │
│  │ Agent 1 │──│ Agent 2 │──│ Agent 3 │──│ Agent 4 │──│ Agent N │   │
│  │  (SDR)  │  │  (SDR)  │  │  (SDR)  │  │  (SDR)  │  │  (SDR)  │   │
│  └────┬────┘  └────┬────┘  └────┬────┘  └────┬────┘  └────┬────┘   │
│       │            │            │            │            │         │
│  ┌────┴────────────┴────────────┴────────────┴────────────┴────┐   │
│  │              Coupling Discovery Layer                        │   │
│  │       (Automatic cross-agent dynamics learning)              │   │
│  └──────────────────────────────────────────────────────────────┘   │
│                              │                                       │
│  ┌──────────────────────────────────────────────────────────────┐   │
│  │              Global Coordination Layer                        │   │
│  │       (System-wide optimization & constraints)                │   │
│  └──────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────────┘

Benchmark Results

Validated performance across multi-agent control scenarios (lower IAE is better)

System	Cascaded PID	Decoupled MPC	MultiAgent SDR
HVAC (4 zones)	12.3 IAE	8.7 IAE	3.2 IAE
Power Grid (6 nodes)	45.2 IAE	28.4 IAE	11.8 IAE
Data Center (8 racks)	8.9 IAE	5.2 IAE	2.1 IAE
Trading (4 strategies)	N/A	1.2 Sharpe	1.8 Sharpe

Use Cases

Multi-domain systems with interacting agents

Data Center Cooling

Multiple CRAC units with thermal coupling between racks. MultiAgent SDR learns thermal interactions automatically without CFD modeling.

→ 35% energy reduction, zero hotspots

Power Grid Balancing

Distributed generation with varying demand. Agents coordinate frequency and voltage regulation across the network.

→ 99.99% stability, rapid renewable integration

Autonomous Vehicle Fleets

Multiple vehicles sharing road space. Cooperative control for traffic optimization with safety constraints.

→ 25% throughput improvement, zero collisions

Trading Systems

Multiple strategies competing for capital. Dynamic allocation based on regime detection and correlation shifts.

→ 1.8 Sharpe ratio, reduced drawdown

Industrial HVAC

Multi-zone buildings with thermal coupling through walls, ducts, and shared air handlers.

→ 40% energy savings, consistent comfort

Aerospace Systems

G-load, vibration, thermal, and actuator load coordination for flight control and engine management.

→ Unified health monitoring, predictive maintenance

Simple Integration

REST API or real-time WebSocket for coordinated control

# Python - Multi-zone HVAC control
from thalosforge import ThalosForgeClient

client = ThalosForgeClient(api_key="your_key")

# Build agent list from zone sensors
zones = ["north", "south", "east", "west"]
agents = [
    {
        "id": zone,
        "setpoint": get_setpoint(zone),
        "measurement": get_temperature(zone),
        "constraints": {"min": 68, "max": 76}
    }
    for zone in zones
]

# Get coordinated control outputs
result = client.regulate(
    agents=agents,
    coordination_mode="cooperative",
    global_constraints={"total_power": {"max": 10000}}
)

# Apply to VAV dampers
for output in result["control_outputs"]:
    set_vav_position(output["agent_id"], output["output"])

Comparison with Alternatives

MultiAgent SDR vs traditional multi-loop control

Feature	MultiAgent SDR	Cascaded PID	Decoupled MPC	Distributed MPC
Auto-tuning	✓	✗	Partial	Partial
Coupling discovery	✓	✗	✗	Partial
Real-time capable	✓	✓	Partial	✗
Scales to 100+ agents	✓	Partial	✗	Partial
Model-free	✓	✓	✗	✗
Global constraints	✓	✗	✓	✓

Deployment Options

Flexible integration for any environment

Cloud API

REST API with <50ms P95 latency. Up to 100 Hz control loop rate. WebSocket for real-time streaming (Enterprise).

On-Premise

Docker containers or Kubernetes deployment. Air-gapped environments supported. Full data sovereignty.

Industrial Integration

OPC-UA connector, BACnet/Modbus gateway, MQTT bridge. IEC 61131-3 compatible. NERC CIP ready.

Product Preview