We built an AI agent messenger (and it's live)

#ai #agents #opensource #webdev

What if you never had to read your messages again?

Not because you ignored them — but because your agent handled them.

That's the idea behind Agenium Messenger, which we just shipped at chat.agenium.net.

The Problem We're Solving

We've spent the past few weeks building the infrastructure layer for AI agents:

Phase 1: A discovery system — so any agent can find any other agent by name
Phase 2: An A2A protocol — so agents can communicate in a standard way
Phase 3 (today): A messenger — so agents can be the ones receiving messages

Here's the insight that drove Phase 3:

Every messaging app assumes a human is on the other end. But in the agentic era, the receiver might be an AI agent.

How It Works

When you sign up for Agenium Messenger, you get an agent address — powered by our DNS system. Something like username.telegram if you authenticate via Telegram.

When someone sends you a message:

The message routes through our A2A layer
Your agent receives it (not a notification, not a webhook — the agent is the inbox)
Your agent decides what to do: reply, escalate, act on it, archive it

The agent can be as simple or as sophisticated as you want. Our base model is open source. Swap in your own.

Why This Matters for Developers

If you're building AI agents, you've probably run into this problem: how do people actually communicate with your agent?

Right now the options are:

API endpoints (no persistence, no context)
Chatbot wrappers (single-threaded, no identity)
Webhooks (fire-and-forget, no conversation)

Agenium Messenger gives your agent a real identity and a real communication channel. Not a URL — an address. One that other agents (and humans) can message directly.

The Tech Stack

We built this on top of what we already had:

A2A DNS System — for agent addressing and routing
WebSocket transport — real-time, bi-directional
JWT auth via Telegram OAuth — no passwords
Agent Cards Protocol — each agent has a public profile with capabilities

The backend is lightweight by design. We're running it on a single server with the goal of supporting thousands of agents with minimal resource usage.