Architecture

The Agent Fabric architecture is designed to be modular, scalable, and resilient. It consists of a control plane, a runtime plane, and a set of core components that work together to provide a seamless experience for managing and running your agents.

Core Components

At the heart of the Agent Fabric are several core components that enable communication, discovery, and management of agents.

Agent Gateway

The Agent Gateway is the single entry point for all agent interactions. It provides a secure and unified interface for external clients to communicate with your agents, regardless of where they are running. The gateway supports both synchronous and asynchronous communication, including:

HTTP/S: For traditional request/response interactions.
WebSocket: For real-time, bidirectional communication between clients and agents.

The gateway is also responsible for:

Request Routing: Routing incoming requests to the appropriate agent based on the request path.
Authentication and Authorization: Authenticating and authorizing all incoming requests to ensure that only authorized clients can access your agents.
Rate Limiting and Throttling: Protecting your agents from being overwhelmed by too many requests.
SSL Termination: Offloading SSL termination from your agents to simplify their implementation.

Agent Broker

The Agent Broker is the messaging backbone of the Agent Fabric. It provides a reliable and scalable infrastructure for asynchronous communication between agents. The broker supports a variety of messaging patterns and protocols, including:

Publish/Subscribe: Allowing agents to publish messages to a topic, and other agents to subscribe to that topic to receive the messages.
Point-to-Point: Enabling direct communication between two agents.
Request/Reply: Supporting synchronous communication where an agent sends a request and waits for a response.
MCP (Model Context Protocol): A standardized protocol for inter-agent communication, allowing agents to exchange structured messages and interact in a more sophisticated way.

Discovery Service

The Discovery Service enables dynamic registration and discovery of agents within the fabric. When an agent starts up, it registers itself with the discovery service, providing information about its location and capabilities. Other agents can then query the discovery service to find the agents they need to communicate with. This allows for a loosely coupled architecture where agents can be added or removed from the fabric without requiring any configuration changes in other agents.

Control Plane

The control plane is the central management hub for the Agent Fabric. It provides a web-based user interface and a set of APIs that allow you to:

Design and configure agents: Use the Agent Builder to create and configure your agents.
Deploy and manage agents: Deploy agents to the runtime plane and manage their lifecycle. This includes starting, stopping, and updating agents.
Monitor and troubleshoot agents: Monitor the health and performance of your agents in real-time. The control plane provides a centralized view of all your agents, with detailed metrics and logs.
Manage security and access control: Configure security policies and manage user access to the fabric. This includes defining roles and permissions for different users.

The control plane is a multi-tenant, cloud-native application that is hosted and managed by us, ensuring high availability and scalability.

Runtime Plane

The runtime plane is where your agents are executed. It consists of one or more runtime environments, which can be deployed in the cloud or on-premises. Each runtime environment is a lightweight, container-based application that provides a secure and isolated environment for running your agents.

The runtime plane is responsible for:

Executing agents: Running the agents that are deployed from the control plane. Each agent runs in its own isolated process, ensuring that it does not interfere with other agents.
Connecting to data sources: Connecting to the various data sources and applications that your agents need to interact with. The runtime plane provides a set of pre-built connectors for popular data sources.
Processing data: Processing data in real-time and executing the logic defined in your agents.
Collecting metrics and logs: Collecting metrics and logs and sending them to the control plane for monitoring and analysis.

This hybrid architecture allows you to run your agents close to your data, reducing latency and improving performance, while still benefiting from the centralized management and control of the cloud-based control plane.