Atomic Agents is an open-source MIT-licensed Python framework by Eigenwise for building AI agent pipelines from small, composable components. Built on top of Instructor and Pydantic, it enforces typed input/output schemas at every step for predictable, testable agent behavior. It supports multiple LLM providers (OpenAI, Anthropic, Groq, Gemini, etc.) and includes a CLI tool called Atomic Assembler for managing reusable components. The framework prioritizes developer control and maintainability over opinionated abstractions. Key features: - Pydantic-based typed schemas for all agent inputs and outputs - Modular components: agents, tools, and context providers that compose cleanly - Multi-provider support: OpenAI, Anthropic, Groq, Gemini, and more - Atomic Assembler CLI for scaffolding and managing reusable agent components - Async support and streaming capabilities - MIT-licensed, 6,000+ GitHub stars, active community
Free and open-source (MIT license).