Package com.paragon.agents

This docs was updated at: 2026-02-23

---

Classs

Name	Description
Agent	A stateful AI agent that can perceive, plan, and act using tools
AgentNetwork	Implements the Network pattern: decentralized peer-to-peer agent communication
AgentResult	The result of an agent interaction, containing the final output and execution metadata
AgentRunState	Serializable state of a paused agent run
AgentService
AgentStream	Streaming agent interaction with full agentic loop
AgenticContext	Holds conversation state for an agent interaction
GuardrailRegistry	Thread-safe global registry for named guardrails
Handoff	A handoff defines when and to which agent control should be transferred
HierarchicalAgents	Implements the Hierarchical pattern: multi-layered supervisor structure
InMemoryMemory	Thread-safe in-memory implementation of Memory with user isolation
InteractableSubAgentTool	Wraps any Interactable as a FunctionTool, enabling composition of multi-agent patterns
MemoryTool	Memory exposed as FunctionTools for agent use
NamedInputGuardrail	An InputGuardrail wrapper that carries a string ID for serialization support
NamedOutputGuardrail	An OutputGuardrail wrapper that carries a string ID for serialization support
NetworkStream	Streaming wrapper for AgentNetwork that provides event callbacks during network discussions
ParallelAgents	Orchestrates parallel execution of multiple agents
ParallelStream	Streaming wrapper for ParallelAgents that provides event callbacks during parallel execution
RouterAgent	A specialized agent for routing inputs to appropriate target agents
RouterStream	Streaming wrapper for RouterAgent that provides event callbacks during routing and execution
SubAgentTool	Wraps an Agent as a FunctionTool, enabling agent composition
SupervisorAgent	Implements the Supervisor pattern: a central agent that coordinates multiple worker agents

Interfaces

Name	Description
GuardrailResult	Result of a guardrail validation check
InputGuardrail	Validates user input before agent processing
Memory	Interface for agent long-term memory storage with user isolation
OutputGuardrail	Validates agent output before returning to the user

Records

Name	Description
AgentDefinition	A richly annotated agent definition record designed for LLM structured output
MemoryEntry	Represents a single memory entry for long-term agent memory
ResponderBlueprint	Serializable descriptor for a Responder configuration
StructuredAgentResult	The result of a structured agent interaction, containing the typed output
ToolExecution	Records the execution details of a single tool call during an agent run