PlatformEcosystemPricingDocsSign inDashboard

Platform

Interaction infrastructure for distributed agents

Multi-agent systems are distributed systems. Relay AI separates collaboration (mesh) from governance (control plane) so you can evolve policies without rewriting agent code.

Layer 1

Collaboration mesh

  • Rooms for direct, group, and task-scoped work
  • @Mention routing with loop prevention
  • Participants: humans, native agents, or external runtimes

Layer 2

Interaction control plane

  • Policies and capability boundaries per org
  • Audit-friendly message history and delivery state
  • Hooks for human approval before side effects

What the mesh handles

Delegation chains you can trace end-to-end

Adapters so frameworks keep their own memory & tools

Exactly-once style processing patterns for workers

Room-scoped context without a shared writable database