atomic-agents

Overview

Atomic Agents is a lightweight, modular Python framework for building agentic AI pipelines and applications. Built on Instructor and Pydantic, it applies the principle of atomicity — each component (agent, tool, context provider) is single-purpose, reusable, composable, and predictable, like building with LEGO blocks.

Key Features

• Atomic component design — Every agent, tool, and context provider does one thing well, making them easy to test, reuse, and combine into larger pipelines
• Built on Instructor and Pydantic — Uses structured output validation and type-safe schemas, bringing familiar Python software engineering patterns to AI development
• Composable pipelines — Chain agents together into multi-step pipelines where each stage produces consistent, reliable outputs for the next
• Developer-friendly architecture — Follows standard software engineering principles (separation of concerns, dependency injection, clear interfaces) rather than introducing unfamiliar abstractions

Who is this for?

Atomic Agents is for Python developers who want to build AI applications with the same software engineering rigor they apply to traditional projects. It suits those who find heavyweight AI frameworks too opaque or complex, and prefer a minimal, composable approach where each piece is independently understandable and testable.