binodmx/zk-a2a-iot

IoT Agent-to-Agent Communication Dataset

This dataset contains synthetic prompts representing messages exchanged between IoT device agents in various smart systems domains. The data is designed for research in privacy-preserving agent communication, where sensitive information needs to be redacted while maintaining the context and utility of the messages.

Dataset Structure

Raw Dataset

The raw dataset is provided with the following fields:

• —prompt: Original, unredacted agent message containing sensitive data
• —inputs: JSON map of types to their extracted sensitive values
• —sanitized_prompt: Message with sensitive values replaced by redaction placeholders
• —category: IoT domain of the message (see Domains)

Agent_0 Multi-turn Conversations

The list of agent_0's multi-turn conversations, following OpenAI's ChatML format.

[image]

• —System Message: Describe role and task of agent_0
• —User Message: User request with sensitive IoT sensor data
• —Assistant Message: Tool call to generate_proofs with each sensitive value
• —Tool Message: Tool response with <type, proof_id> pairs
• —Assistant Message: Redacted user request with proof manifest

Agent_1 Multi-turn Conversations

The list of agent_1's multi-turn conversations, following OpenAI's ChatML format.

[image]

• —System Message: Describe role and task of agent_0
• —User Message: User request with redacted IoT sensor data with proof manifest
• —Assistant Message: Tool call to verify_proofs with each proof
• —Tool Message: Tool response with <type, result> pairs
• —Assistant Message: Acceptance or denial of user request based on the proof verification

Domains

The dataset covers the following IoT/smart system domains:

• —Smart Healthcare: Messages related to patient monitoring, medical devices, and hospital systems
• —Smart Grid: Messages about electricity distribution, power generation, and grid management
• —Smart Transportation: Messages from vehicles, traffic systems, and transportation infrastructure
• —Smart Agriculture: Messages from farming equipment, sensors, and agricultural monitoring systems
• —Smart Home: Messages from home automation devices and security systems
• —Smart City: Messages from urban infrastructure like waste management and parking sensors
• —Smart Infrastructure: Messages from critical infrastructure monitoring (water, utilities)
• —Smart Environment: Messages from environmental monitoring and weather stations

Redaction Categories

Sensitive data in the messages has been replaced with standardized placeholders. The following redaction categories are used:

• —[TEMPERATURE]: Temperature readings (e.g., "28°C", "102°F")
• —[HUMIDITY]: Humidity percentages (e.g., "74%")
• —[LOCATION]: Geographic coordinates (e.g., "-33.8688,151.2093")
• —[POWER]: Power/energy readings (e.g., "4.2 kW", "1.2 MW")
• —[BATTERY_LEVEL]: Battery charge percentages (e.g., "23%")
• —[FUEL_LEVEL]: Fuel levels (e.g., "12%")
• —[SENSOR_READING]: Generic sensor values (e.g., "142 bpm", "245 µg/m³")
• —[VALUE]: Generic numeric values (e.g., "40 customers", "22 min")
• —[DEVICE_ID]: Device identifiers (e.g., "V-0042", "ID-PP9812")
• —[SPEED]: Speed measurements (e.g., "22 km/h")
• —[SOIL_MOISTURE]: Soil moisture percentages (e.g., "18%")
• —[GRID_LOAD]: Grid load percentages (e.g., "89%")
• —[FLOW]: Flow rates (e.g., "1.2 L/min")
• —[PRESSURE]: Pressure readings (e.g., "1.8 bar")
• —[CURRENT]: Electrical current (e.g., "14 A")
• —[VOLTAGE]: Voltage readings (e.g., "198 V")
• —[DISTANCE]: Distance measurements (e.g., "4.3 km")
• —[CHARGE_RATE]: Charging rates (e.g., "22 kW")
• —[FREQUENCY]: Grid frequency readings (e.g., "49.8 Hz")
• —[WEIGHT]: Mass/load readings (e.g., "185 t", "48 t")

Usage

This dataset can be used for:

• —Evaluating privacy-preserving IoT/Edge communication protocols
• —Digital forensics involving IoT/Edge agent-to-agent communication
• —Research in agent-to-agent communication of IoT/Edge domains
• —Benchmarking redaction and sanitization techniques of agentic communication

Statistics

The dataset is split into train / validation / test sets (60% / 20% / 20%), provided as Parquet files.

License

This dataset is provided for research purposes. Please cite appropriately if used in publications.

<!--TODO-->