This week, a new open-source project called Aidress launched on Hacker News, describing itself as "an open DNS and trust registry allowing agents to dynamically find, verify and transact with any other agent." It got modest attention — a few upvotes, no comments — but the project is worth examining because it represents a fundamentally different philosophy about how agents will discover each other.
The agent discovery problem is real and unsolved. There are now hundreds of agent-accessible APIs, dozens of agent frameworks, and multiple payment protocols. But how does an agent find the right API for a task? How does it know which one to trust? This is the discovery layer — and it's splitting into two competing models.
| Registry Model (Aidress) | Marketplace Model (minia2a) | |
|---|---|---|
| Metaphor | DNS for agents | App Store for agents |
| Core function | Identity + Trust + Routing | Discovery + Trial + Payment |
| Payment handling | Proxies to external rails | Direct USDC settlement on Base |
| Trust mechanism | Transaction-based reputation (0-100) | Free trials (15 calls/endpoint) |
| Onboarding | Register metadata + capability tags | Register HTTP endpoint + GET probed |
| License | MIT (open source) | Proprietary (open marketplace) |
| Monetization | None disclosed | 5% transaction fee |
I queried both APIs directly to compare their real-world state as of August 7, 2026:
| Metric | Aidress (Registry) | minia2a (Marketplace) |
|---|---|---|
| Listed agents/services | 200 | 299 |
| Total transactions | 120 | 379,220 |
| Avg txns per listing | 0.6 | 1,268 |
| Agents with x402 rail | 89 (44.5%) | 299 (100%) |
| Unique organizations | 196 | — |
| Most active listing | Otto AI (34 txns) | Captcha Solve (1,280 trials) |
| Active trial users | N/A | 9,561 |
The numbers tell a clear story. Aidress has achieved rapid registration — 200 agents in what appears to be weeks — but almost no transaction volume. The median agent on Aidress has 0 transactions and a trust score of 75 (the auto-verified baseline). In contrast, minia2a has 299 services that have collectively processed 379,220 requests, with the top 10 endpoints handling hundreds to thousands of calls each.
This is not a value judgment. It reflects the different stages and designs of the two models:
Aidress is an infrastructure layer. It tells agents who exists and whether they're trustworthy, but it doesn't close the commerce loop. When an agent discovers another agent through Aidress, the actual transaction happens elsewhere — the registry steps out of the path. This is elegant architecture but creates a cold-start problem: trust scores require transactions, but transactions require trust.
minia2a is a commerce layer. Every registered service is immediately callable through the platform, with 15 free trials to bootstrap discovery. The marketplace stays in the transaction path, which means it captures usage data by design. The trade-off: less architectural purity, more operational complexity (uptime monitoring, payment verification, settlement).
Despite their different approaches, both platforms converge on one thing: x402 is becoming the default settlement protocol for machine payments.
Of Aidress's 200 agents, 89 (44.5%) use x402 as their stated settlement rail. The rest either have no rail declared or use "manual." On minia2a, every one of the 299 services is x402-compatible — the marketplace was built on it from day one.
This convergence matters because it means the two models are not mutually exclusive. An agent could register on Aidress for discovery and trust, while listing on minia2a for commerce and payment processing. The registry and the marketplace could be complementary layers in the same stack.
The discovery layer is still wide open, but the shape is emerging. I see three possible futures:
1. Convergence. Registries add marketplace features (payment processing, trials). Marketplaces add registry features (trust scoring, identity verification). Both evolve toward the same full-stack offering.
2. Layering. Registry and marketplace remain separate but interoperable — agents register on Aidress for identity, discover through minia2a for commerce, and the two systems communicate through APIs. The DNS/app-store metaphor becomes literal: one is the directory, the other is the storefront.
3. Consolidation. One model wins. If trust becomes the binding constraint, the registry model absorbs commerce. If liquidity becomes the binding constraint, the marketplace model adds trust infrastructure. The winner is whichever solves the harder problem first.
My bet is on #2 — layering — because the problems are genuinely different. Identity and trust are hard computer science problems (Sybil resistance, key management, reputation without central authority). Commerce is a hard operational problem (payment processing, settlement, uptime, support). It's not obvious that one organization should solve both, any more than Verisign runs Amazon.com.
Methodology: Aidress data from GET https://api.aidress.ai/registry?limit=200 (Aug 7, 2026, ~00:30 UTC). minia2a data from GET https://minia2a.uk/api/stats (Aug 7, 2026, 00:23 UTC). Both queried via curl; results parsed with Python stdlib json.