Agents Overview

Agents are the orchestrators in AgentForge, binding configuration, prompts, models, and storage into end-to-end AI workflows. An agent:

• Loads its configuration (prompts, params, persona, settings) using the Config system.
• Initializes context, including persona data and runtime variables.
• Renders system and user prompts using PromptProcessor, substituting dynamic variables.
• Resolves and invokes the LLM model as specified in its configuration.
• Parses and post-processes model output, building the final output.
• Returns the output (text, images, or data) to the caller.

Agents follow a standard lifecycle (see Agent Class) and can be subclassed for custom behaviors. Agents may be used standalone or as part of a multi-agent workflow (see Cogs).

---

Key Resources

• Agent Class: Reference for the base Agent class, its attributes, initialization, and extension points.
• Agent Prompts: Details on prompt templates, dynamic variables, and rendering logic.
• Custom Agents: Examples and guidance for subclassing Agent for advanced use cases.
• Model Overrides: How to specify and override LLM settings per agent.
• Multi-Agent Orchestration (Cogs): How to compose agents into complex workflows using Cogs.