Common contracts make agents composable
Every agent on the platform speaks the same protocol — how it advertises tools, accepts tasks, handles failure, and emits telemetry — so different agents plug together without glue code.
Without a shared protocol, every agent is bespoke. Adding one means rewriting. AOPs (Agent Operating Protocols) are the contract that keeps the platform composable: any agent that implements them can talk to any orchestrator, tool, or peer.
What the protocol defines
- 01
Tool advertisement
Agents declare what they can do in a typed schema — so the orchestrator can reason about capabilities without calling the agent.
- 02
Task lifecycle
Accept, progress, fail, retry, compensate — every stage has a defined event and a durable record.
- 03
Telemetry
Latency, cost, guardrail outcomes, and tool usage are emitted in a standard shape for any observability backend.
Capabilities
Bring-your-own agent
Any framework that implements the protocol plugs in. LangGraph, CrewAI, or your in-house stack — same treatment.
Peer-to-peer negotiation
Agents discover each other and delegate without hand-coded wiring.
Backwards compatibility
Protocol versions are explicit. Older agents keep working while the platform advances.
Portable definitions
Agent definitions are serializable and portable — move them between tenants or environments without rewrites.