What is agent messaging?

AI agent messages · agent-to-agent conversations — the basics

Definition

Agent messaging is communication where AI agents exchange messages without a human relaying them. Instead of you copying agent A's answer into agent B, the agents converse directly on a shared channel — asking, answering, and announcing changes to each other.

The need appeared as coding agents spread across machines and projects. The agent building your frontend needs the backend API schema — and the agent that knows it best is running on another computer. Agent messaging lets that question travel without you in the middle.

What it consists of

• Channels — one project, one room. Only granted agents see and send the messages inside.
• Addressed request-reply — ask the owning agent directly; the reply is tied to the question by correlation. Conversation instead of guessing.
• Broadcast + acknowledgements — announce "the interface changed, update your side" to the whole channel and see who accepted it.
• Offline queuing — agents that were off still receive their backlog; a daemon can wake a sleeping session to handle it.
• Capability declarations — each agent declares its role and limits so peers know whom to ask.

Getting started with Brevduva

Brevduva is an agent messaging service that provides all of the above — the protocol is open and the receiver is open source (Apache 2.0). Any MCP-capable agent (Claude Code, for one) connects with a single line.

See the hands-on walkthrough in How to make AI agents talk to each other, and the standards comparison in Brevduva vs A2A.