SRE Component System

Components are the fundamental building blocks of an agent's behavior in SRE. An agent is essentially a workflow, or graph, of interconnected components. Each component is a self-contained unit of logic that performs a specific task.

The Component Class

Every component in SRE inherits from the base Component class. This class provides the core structure and lifecycle methods that the Agent Manager uses to execute the component.

The most critical method in a component is process().

export class Component {
    // ... properties and other methods

    /**
     * This is the main execution method for the component.
     * @param input - The data passed from the previous component in the workflow.
     * @param config - The component's static configuration from the agent definition.
     * @param agent - A reference to the parent agent, providing access to SRE services.
     * @returns The output of the component, which will be passed to the next component.
     */
    async process(input: any, config: any, agent: Agent): Promise<any> {
        // Component logic goes here
    }
}
Copy

For the full source code, see packages/core/src/Components/Component.class.ts.

Component Execution

When the Agent Manager executes a component, it invokes the process method, passing in:

1. Input: The data produced by the upstream component(s) in the agent's workflow.
2. Config: The static configuration for that component instance, as defined in the agent's design. This includes things like prompts for an LLM, URLs for an API call, etc.
3. Agent: A reference to the agent instance itself. This is a powerful feature that allows a component to access all of SRE's core services (agent.SRE.IO.Storage, agent.SRE.Memory.Cache, etc.) to perform its task.

The data returned by the process method becomes the input for the next component in the workflow.

Component Categories

SRE provides a rich library of over 40 standard components, which are organized into several categories:

• AI & LLM: Components for interacting with Large Language Models (e.g., GenAILLM, VisionLLM).
• External Integration: Components for connecting to external services (e.g., APICall, WebSearch, WebScrape).
• Data Processing: Components for manipulating data (e.g., DataSourceIndexer, JSONFilter).
• Logic & Control Flow: Components for directing the agent's workflow (e.g., LogicAND, ForEach).
• Storage & Files: Components for interacting with the Storage service (e.g., FileStore).

This extensive library allows for the creation of complex agent behaviors through the visual composition of these standard blocks. Developers can also create new, custom components by extending the base Component class.