A2A protocol

Agent2Agent (A2A) Protocol is an open standard for independently operated agent systems to discover capabilities and exchange messages, tasks, status updates, and artifacts without exposing their internal models, memory, or tools. It defines an optional interoperability boundary, not the queue or message bus inside either agent system. A system does not need A2A merely because it contains multiple agents.

What the protocol defines

A2A covers capability discovery through Agent Cards, authenticated message exchange, task lifecycle, status updates, artifacts, streaming, and asynchronous delivery. The current specification provides equivalent protocol bindings for JSON-RPC, gRPC, and HTTP with JSON.

Where eqlx uses it

A2A belongs at a deployment boundary when an agent system eqlx builds and the customer runs must work with an independently deployed agent system. The requesting side discovers the remote capability, authenticates, submits or streams a message, and receives task status and artifacts. Each side remains opaque and continues to own its internal policy and execution.

Inside a system eqlx builds, orchestration may use a durable event log or task queue for typed handoffs, retries, ordering, replay, and backpressure. That internal event layer has its own schema and operational guarantees. It is not called A2A unless the boundary actually implements the A2A specification.

Bounded execution still belongs to the orchestrator

A2A does not decide how many times a specialist may retry, which actions need approval, or when a run must stop. The orchestrator sets a deadline and retry budget before delegation. A completed artifact produces a terminal receipt; an exhausted budget routes to a deterministic fallback, a named human-review queue, or a failed state. Protocol interoperability does not replace operational accountability.