ReadyForAIDigital workforce infrastructureBook a demo

Multi-Agent Collaboration

How Agents from different frameworks are governed under one roof, which protocols they speak, and how they split work as Coordinator + Workers while handing decisions that need a human back to a human.

Heterogeneous Agents can split work as Coordinator + Workers without unifying Agent technology

  • NodalOS governs heterogeneous Agents through five RuntimeClasses and connects them across frameworks via Fabric and A2A / ACP / MCP
  • The Coordinator orchestrates and specialist Workers execute; each side gets the context it needs from HarnessServer
  • PathPilot handles dispatch, tracking, and escalation, and resource locks keep Agents from editing the same resource at once
  • Decisions that need a human and stuck tasks land in the AULO inbox; human decisions are written back upstream
Agents on different frameworks cannot talkNo shared identity or messaging protocolWhen a chained task fails, nobody knows whereStuck tasks never reach a humanSeveral Agents edit the same resource at once
Heterogeneous HostingNodalOS

Five RuntimeClasses — Native / Vendor / Sidecar / Basic / Coordinator — with lifecycles managed by the agentd daemon

Messaging ProtocolsNodalOS

Five standardized Fabric messaging mechanisms with native A2A / ACP / MCP interoperability; multi-product MCP aggregated by namespace

Coordinator + WorkersNodalOS

The Coordinator orchestrates while specialist Workers execute; the platform ships with a Coordinator and a Basic Agent

Dual-Side ContextHarnessServer

Workers pull a task context index when work starts, Coordinators pull SOPs per action signal, and each side sees different content by visibility

Task Control PlanePathPilot

Dispatch, tracking, blockage handling, and human escalation; can recommend an executor by skill and occupancy scoring

Human–Agent CollaborationAULO

One-on-one chat with the Coordinator; escalations, HITL approvals, and pending council topics land in one inbox

  1. 01The owner chats with the Coordinator in AULO
  2. 02Tasks are dispatched in PathPilot, with an executor recommended
  3. 03Workers pull task context and start
  4. 04Agents collaborate over Fabric / A2A
  5. 05Progress is self-reported; blockers escalate
  6. 06The owner decides in the AULO inbox

Every step in the chain has an owner — orchestration sits with the Coordinator, execution with Workers, task state with PathPilot, and anything that needs a human decision goes to the AULO inbox.

ProblemProductCapability
Claude Code, Codex, and in-house Agents coexist but cannot interoperateNodalOSFive RuntimeClasses under unified management, interoperating over A2A / ACP / MCP
Messages and calls between Agents follow no conventionNodalOSFive standardized Fabric messaging mechanisms for cross-framework interoperability
Orchestration and execution are tangled in one AgentNodalOSCoordinator topology orchestration, with every Agent's lifecycle managed by the agentd daemon
When chained Agents fail, nobody can tell at which stepPathPilotActionable pipelines — multi-stage flow, decision gates, reset / skip — with self-reported progress collected on the execution timeline
Several Agents edit the same resource at oncePathPilotResource locks hold a shared resource across tasks to avoid concurrent conflicts
Stuck tasks never reach a humanAULOEscalations land in the unified inbox, where humans decide and results are written back upstream

By role

Technology leaders

Five RuntimeClasses plus A2A / ACP / MCP, with no lock-in to a specific framework or model vendor

Compliance & risk

High-risk actions intercepted by OwlAudit policies go to human review, with approvals and releases recorded in AULO

Business leaders

People focus on goals and key trade-offs, the Coordinator handles orchestration, and decisions that need a human land in the inbox

What else do you want to know?

Harness EngineeringSecurity & PermissionsExisting architecture integrationBook a demo
0 / 500
SYSTEM READYintent/multi-agent-collaboration