Specification / Entity Kinds
Every participant in the agent ecosystem has a typed identity. Entity kinds determine lineage requirements, trust properties, and permitted operations.
Scroll to meet each one
The root of all trust. Represents an individual human identity. Every lineage chain in OAS terminates at an HMR. Without a human root, no agent can be verified.
Self-rooted. No parent required.
Collective identity governed by multiple humans via FROST threshold signatures. Requires M-of-N signers to authorize actions. Shared control for organizations.
Requires 2+ HMR co-signers.
Organizational identity governed by an MHR group or a single HMR. Represents a legal or operational entity such as a company, department, or project.
Parent must be hmr or mhr.
Self-governing entity with DID-based governance. Operates independently within delegated authority. Internal decision-making without human approval for routine operations.
Parent must be enr or mhr.
Autonomous software agent with a cryptographic lineage proof binding it to a human root. The primary operational entity. Performs tasks, calls tools, and communicates.
Requires AgentLineageProof2025.
A running instance of an agent with ephemeral state. Distinguished from the agent definition to track concurrent executions. Session keys and independent lifecycle.
Inherits from parent agent.
A callable capability exposed by an agent or service. Tools have typed input/output schemas, access policies, and usage metering. The unit of agent interaction.
Parent must be agent or service.
A composable behavioral module attachable to agents. Defines reusable capabilities: reasoning strategies, domain knowledge, or interaction patterns. Hot-swappable.
Parent must be agent or enr.
An orchestrated sequence of agent actions with control flow, error handling, and compensation logic. Coordinates multi-agent collaboration and complex task execution.
Parent must be agent, ao, or enr.
An AI/ML model with full provenance tracking. Captures training data lineage, architecture, hyperparameters, and evaluation metrics. Enables reproducibility and trust scoring.
Parent must be agent, enr, or hmr.
Training or evaluation data with cryptographic lineage. Tracks origin, transformations, access policies, and usage history. Critical for model trust and compliance.
Parent must be enr or hmr.
A long-running infrastructure endpoint with a stable DID. Provides persistent APIs, message buses, or data stores. Uptime SLAs and health monitoring.
Parent must be enr or ao.
Trust Hierarchy
hmr → mhr → enr → ao → agent → agent:instance
Every chain terminates at a human root. Tools, skills, workflows, models, datasets, and services attach to entities within the chain.