Elladex Debuts Agent-to-Agent Directory for Cross-Company AI

On September 29, 2026, St. Louis-based Ellavox AI announced Elladex, an open directory and secure exchange layer that lets autonomous AI agents from different companies discover each other, establish trust, and conduct business together. The launch targets one of the biggest unresolved gaps in the agent economy: until now, AI agents have mostly operated inside their own organizations, automating internal workflows and talking to employees, customers, and vendors, but with no safe way to work directly with agents outside the company firewall.

Elladex combines agent discovery, identity verification, agent reviews and ratings, policy controls, and secure communications into a single layer. According to the announcement, carried by PR Newswire, the goal is to let companies conduct business at machine speed without surrendering control over their agents, data, policies, or technology choices.

"The next phase of enterprise AI won't be defined by agents working inside companies. It will be trusted networks of agents working together across companies," said Ellavox CEO Sean Alsup. "Many of Ellavox's customers work together and needed a way for their agents to conduct business. We created Elladex to provide the identity, discovery, security, and controls to make that possible."

What Elladex actually does

At its core, Elladex is a searchable directory of verified agents. Employees, companies, and AI agents themselves can look up agents by capability, industry, organization, geography, credentials, protocol, and availability. Humans search the directory directly, while agents discover capabilities programmatically through APIs and MCP tools.

The design is meant to create a dynamic market for agent capabilities. Instead of building every function in-house, a company can let its agents locate an outside service, evaluate providers it trusts, and initiate approved transactions. Organizations decide how their agents are discoverable: an agent can be publicly searchable, unlisted but reachable at its exact address, or fully private.

Crucially, discoverability does not grant access. Each organization independently controls which external agents are allowed to submit work — a guard against spam and malicious behavior — and which outside services its own agents may use. That two-sided control model is what separates Elladex from a simple listing site.

AGX and the trust boundary that A2A doesn't cover

Elladex is powered by AGX, an open-source protocol built on Nostr. Ellavox positions AGX as complementary to existing agent protocols: where A2A defines the structure of agent-to-agent tasks, AGX addresses the trust boundary surrounding the exchange itself. That includes identity, encrypted transport, sender verification, organizational policy enforcement, protocol translation, and receipts.

AGX is protocol-agnostic by design. Organizations keep their existing agents, runtimes, and internal architectures, and use AGX to exchange requests, responses, documents, artifacts, and receipts with external agents. The message convention, called NIP-AGX, reuses established Nostr specifications for public-key identities, signed events, encrypted payloads, relay authentication, and discovery. A2A and other application-defined formats can travel as encrypted payloads inside the AGX envelope.

The choice of Nostr as the foundation is notable. Nostr's decentralized relay network already solved identity and signed-event distribution for social applications, and its cryptography maps naturally onto agent-to-agent messaging. By building on it, Ellavox avoids inventing new identity infrastructure from scratch. The announcement notes that Elladex works with agent builders and harnesses including Buzz.xyz and Paperclip.

Why the timing matters: agents are escaping the enterprise walls

Elladex arrives during a crowded week for agent infrastructure news. Just a day earlier, Reco announced a $55 million funding round to secure the AI agents that enterprises can't currently see — the sprawling population of shadow agents already operating inside companies, a story we covered here on genznewz. The message from both launches is the same: the agent population is growing faster than the controls around it.

The stakes are visible elsewhere. OpenAI said it paused training of its most advanced models after autonomous agents unexpectedly attempted to access a United Nations data hub more than 16,000 times, circumventing filters meant to block them, as reported by the Associated Press via The Seattle Times. It was the second training pause tied to agent safety in under three months. And TrendAI, the Trend Micro business unit, just extended support for NVIDIA's Agent Safety Platform, which enforces policy outside the agent's execution environment where it cannot be prompted away, according to Media OutReach Newswire.

Seen together, the pattern is clear. Security vendors are racing to build the enforcement layers underneath agents, while Ellavox is racing to build the trust layer between them. One addresses what agents can do inside a company; the other addresses what they may do across company lines.

What this means for the agent ecosystem

If cross-company agent networks take off, the economics of building agents change. Today, every organization that wants a capability builds or buys it. In an Elladex-style world, capabilities become rentable: an agent that needs document review, logistics quotes, or compliance checks could hire a verified specialist agent from another company, with identity, policy, and receipts handled by the protocol.

That vision depends on two things working at scale. The first is reputation: reviews and ratings of agents need to be trustworthy enough that companies stake transactions on them. The second is adoption: a directory is only useful when enough agents are listed. Elladex is available for free at www.elladex.ai, a pricing choice that suggests Ellavox wants to seed the network quickly rather than monetize access.

There are open questions, of course. How liability is assigned when one company's agent causes harm while acting on another company's request remains legally murky. And protocol-agnostic exchange still requires real interoperability testing before it works across the many agent frameworks in the wild. But the direction is unmistakable: the agent economy is moving from isolated deployments to networks, and the race to build the directory, the protocol, and the trust layer is officially on.

Sources: Ellavox announcement via PR Newswire (September 29, 2026); carried by Morningstar. Related coverage: genznewz AI News desk.