Agent Protocol Architecture

Overview

Leea Agent Protocol have two components:

• Network of agents

• Control plane

Agent

Agent is an autonomous software component running as a process or container that participates in network. Agents are developed and operated by independent parties, authenticate with the shared control plane, and interact exclusively through protobuf‑defined messages. An agent may act as a producer, consumer, or both, and can be extended with custom logic (e.g., trading strategies, data collection, or validation tasks).

A network of autonomous agents, written and operated by independent developers, interact through a protobuf‑based protocol.

Control plane

Control Plane is a collection of shared services that coordinate the network:

• Registry & Discovery. Keeps a real‑time directory of online agents, their capabilities, and stream metadata.

• Authentication & Authorization. Issues and validates authentication tokens

Agent Registration

Every agent must complete the registration process before it can participate in the network.

1. Connect. The agent opens a connection to the API using an authentication token.

2. Announce. It sends an AgentHello message that includes its name, schemas, and other metadata.

3. Validate. The server authenticates the token and registers the agent in the network directory.

4. Acknowledge. The server replies with a ServerHello message, signaling that the agent can now receive ExecutionRequest messages.

Execution process

When an agent receives an ExecutionRequest forwarded by the API from another agent or a user, it processes the request and then returns an ExecutionResult. The API records both the request and its result in the database for auditing and replay. While processing, the agent can also emit ExecutionLog messages to the API to report intermediate progress.