Skip to main content

Agent-to-Agent — A2A

This page describes multi-agent patterns (“agent-to-agent”) in ARP terms.

WIP

ARP Standard v1 does not define a dedicated A2A protocol, and the first-party JARVIS stack does not ship a multi-agent coordination layer today. The patterns below are guidance for integrating A2A behavior into the node-centric execution fabric.

Pattern 1: “Agents as capabilities” (atomic NodeTypes)​

Expose “call another agent” as one or more atomic NodeTypes. The handler:

  • sends a request to the remote agent API
  • normalizes the response into typed outputs
  • returns those outputs as the atomic NodeRun outputs

This keeps the execution fabric unchanged: planners/selectors see the remote agent as a capability option, and policy can gate it like any other node type.

Pattern 2: Orchestrator at the application layer​

Have your application orchestrate multiple runs explicitly:

  • run Agent A → read result
  • feed result into Agent B as context/data
  • aggregate results in your own “controller” logic

This is often the simplest approach when custom routing or business-specific policies are needed, but may involve ARP Standard Contract change/addition.