Skip to main content

Architecture Overview

AutoAgents is built with a modular, extensible architecture that prioritizes performance, safety, and developer experience. This document provides a comprehensive overview of the framework's design and core components.

High-Level Architecture​

Key layers:

  • Agent Definition: your agent’s metadata, tools, and output
  • Executors: Basic (single‑turn), ReAct (multi‑turn with direct tool calls), and CodeAct (multi‑turn with sandboxed TypeScript tool composition)
  • Memory: context storage (e.g., sliding window)
  • Tools/MCP: capabilities the agent can call
  • Runtime: optional actor system for multi‑agent workflows
  • Providers: pluggable LLM backends (cloud/local)

In practice, execution starts with an agent definition. That definition is paired with an executor, which decides how the task is handled, whether the run stays single-turn or enters a multi-step reasoning loop with tool calls. During a run, the executor can read from memory, invoke tools or MCP-backed capabilities, and dispatch prompts to an LLM provider.

Direct agents execute this flow inline and return results to the caller. Actor agents add the runtime layer, which lets multiple agents communicate through topics, coordinate background work, and participate in larger workflows. The runtime sits above the provider layer, so the same agent logic can target cloud or local models without changing the core execution model.

This separation is what keeps AutoAgents modular: agent behavior, execution strategy, memory, tools, runtime orchestration, and model providers can evolve independently while still composing into one consistent system.