ORBIS

Appearance

Agent-to-agent (A2A) ​

When ORBIS hands work to one of your delegates, it might be talking to another agent — not just calling a model. The way it does that is A2A, the Agent2Agent protocol. This page explains what A2A is and why it lets ORBIS plug into a whole fleet of agents.

What A2A is ​

A2A is an open, standard way for AI agents to talk to each other — a JSON-RPC contract over HTTP for sending a message to an agent, holding a conversation across turns, and streaming progress back. It's the "USB-C for agents": any agent that speaks A2A can work with any client that speaks A2A, with no bespoke integration.

ORBIS is an A2A client. Add an A2A delegate, and the orb knows how to send it a request and interpret what it sends back.

Why it matters ​

Because A2A is a standard, any A2A agent becomes a delegate — drop in its URL and a description, and the orb can route work to it. ORBIS stays a fast conductor; the specialists in your fleet do the deep work. No glue code per agent.

What you get over a plain model call ​

An OpenAI-compatible delegate is a single, stateless model call. An A2A delegate is a full agent interaction:

  • It remembers the thread. ORBIS keeps a conversation handle with each A2A delegate, so you can follow up — "and dig into that incident" — and the delegate has the context.
  • It streams progress. A long task can report as it goes; the orb narrates that progress so you're not sitting in silence.
  • It has a lifecycle. A delegate can finish, fail, or come back asking for more input — ORBIS handles each: it answers from context if it can, or surfaces the question to you.
  • It describes itself. Agents advertise their capabilities (streaming, multi-turn) via an "agent card"; ORBIS adapts to what each one supports.

How a hand-off flows ​

  1. You ask for something that fits a delegate's description.
  2. The orb acknowledges immediately and sends the request over A2A.
  3. The delegate works — streaming progress if it can.
  4. The orb speaks the synthesized result, attributed to the delegate.

For multi-step asks, ORBIS can orchestrate several A2A delegates in the background — see The agent model.

See also ​