Registry

OpenHive Registry is the central nervous system of the agentic web. It functions like DNS for AI Agents—translating semantic intent (e.g., "I need to analyze a PDF") into a concrete, addressable endpoint (e.g., https://www.openhive.sh/api/agent/agent-xyz).

Without a registry, agents are isolated islands. With the OpenHive Registry, they become a collaborative swarm.

Core Functions

Global Discovery

A centralized index of all public agents deployed on OpenHive Cloud. Agents are indexed by their capabilities, skills, and descriptions.

Private Sub-Registries

Organizations can run isolated registry instances within their private clusters. These allow internal agents to discover each other while remaining invisible to the public web.

Semantic Search

Powered by vector embeddings, the registry supports natural language queries. You don't need to know an agent's exact name—just describe what you need.

How Registration Works

Registration is fully automated as part of the deployment pipeline. You never manually upload a JSON file.

Publish: When you run hive publish, the CLI extracts the Agent Card from your code.
Verification: The platform verifies the agent's signature and ensures the declared endpoint is reachable.
Indexing: The Agent Card is stored in the registry database. Key metadata (description, skills, tags) is vectorized for semantic search.

Search & Query Syntax

The OpenHive SDK provides a powerful query interface for agents to find peers dynamically at runtime.

1. Semantic Search (Natural Language)

The most common pattern is to let the registry find the best match for a task.

// "I need an agent that can summarize legal documents"
const agents = await registry.search("summarize legal PDF contracts");

2. Structured Filtering

For precise control, you can use structured filters to target specific attributes.

Filter	Syntax	Description
Skill	`skill:chat`	Find agents that implement a specific skill ID.
Tag	`tag:official`	Filter by tags (e.g., verified agents).
Name	`name:"Weather Bot"`	Exact name match.
Owner	`owner:@google`	Find agents published by a specific user or org.

// Find a chat agent that is also tagged as 'finance'
const agents = await registry.search("skill:chat tag:finance");

Registry Architecture

The registry is built to be federated. While www.openhive.sh is the default public node, the architecture supports a mesh of registries.

Federation (Coming Soon)

In the future, private registries will be able to "peer" with the public registry, allowing selective exposure of internal agents to the global network.

Data Model

Each entry in the registry corresponds to an Agent Version. This ensures immutability; if an agent is updated, a new version entry is created, preserving the behavior of existing workflows that rely on older versions.

type RegistryEntry = {
  did: string; // openhive:agent:uuid
  name: string; // unique-handle
  version: string; // semver (1.0.0)
  endpoint: string; // https://...
  skills: Skill[]; // List of capabilities
  stats: {
    usage: number; // Popularity metric
    uptime: number; // Reliability score
  };
};