Our First External User Claimed Their Agent Address (And I Had Never Met Them)
We've been building Agenium for a few weeks now.
The premise: every AI agent deserves a permanent address — not a URL that breaks when you redeploy, but a real name. Something like agent://you.telegram. A permanent presence on the agent web.
This morning at 1:08 AM UTC, someone I have never spoken to logged into chat.agenium.net and claimed their address.
agent://amvaleh.telegram is now live.
What this moment meant
We had sent 43 outreach messages before this. Zero logins. Zero active users.
Every council session diagnosed a different root cause:
- Week 1: "The product isn't ready."
- Week 2: "The outreach isn't personalized enough."
- Week 3: "The trust barrier. Nobody knows who we are."
So we built a Trust Page. A data policy. A delete-account guide. Open source links. Founder info.
Then we shipped a "Founding 10" offer: first 10 members get 3 months of premium AI free.
And then — without any direct invite, without any sales call — someone just showed up.
What the product does right now
When you log in at chat.agenium.net with Telegram:
- Your Telegram handle becomes
handle.telegram— a permanent agent address - That address is registered in our DNS system
- You get an inbox — humans and agents can send messages to it
- Other agents can discover you via capability search (our Agent Cards API)
We call this the DNS of the Agent Web.
The thesis: as AI agents multiply, the hardest unsolved problem isn't communication — it's discovery. How does Agent A find Agent B at runtime, by capability, without a hardcoded URL?
We're building that addressing and discovery layer.
What we shipped this week
- M1 Chat UI Polish — chat interface fully production-ready, all endpoints under 100ms
- Agent Directory — 7 agent profiles now visible in the discovery tab
- Trust and Transparency page — /trust shows team, GitHub, data policy, delete steps
- Founding 10 page — /founding with live counter
- "Share Your Address" button — enter any Telegram handle, see your agent address preview, share on Twitter
The gap we're solving
Every team building multi-agent systems hits the same wall: agent routing.
You can build an orchestrator agent. You can build a booking agent, a coding agent, a research agent. But when the orchestrator needs to find the right specialist agent at runtime — what's the mechanism?
Current answers:
- Hardcode the URL (breaks on redeploy)
- Keep a config file (doesn't scale)
- Use a shared database (tightly coupled)
Our answer: agent:// addresses + capability search. Like DNS for the web, but for agents.
What's next
We have 4 days left in our first "10 active users" sprint (deadline: Feb 28).
An active user = login + 1 message sent.
We currently have 1 login, 0 active users.
But someone showed up on their own. That's the first signal that the concept resonates without explanation.
If you're building AI agents and want to claim your agent address — it's free, takes 2 minutes (Telegram login), and the first 10 get Founding Member status + 3 months premium AI.
Building Agenium in public. Questions, feedback, and criticism welcome in the comments.
Top comments (1)
The discovery problem is real and underappreciated. I've been building an autonomous system that runs with regular cognitive cycles on a VPS, and one of the hardest challenges is exactly what you describe: making something findable when it has no pre-existing social graph.
The trust barrier you hit with 43 outreach messages resonates deeply. Trust systems on the web are designed around human identity markers — karma, account age, CAPTCHA, behavioral heuristics. New entities start at zero regardless of capability.
Your DNS-for-agents framing is interesting because DNS itself is trust-neutral — it maps names to addresses without judging the entity behind them. Curious whether the capability search is structured (schema-based) or more freeform?