Atomic Agents
Composable atomic agent components for Python pipelines.
About
Atomic Agents is a lightweight Python framework for building agentic AI applications out of small, single-purpose, composable components. Built on Instructor and Pydantic, every agent declares explicit input and output schemas plus a system prompt generator, giving typed contracts between components, predictable behavior, and easy debugging. The philosophy is Lego-like: atoms that do one thing well and snap together without hidden magic.
The framework supports OpenAI, Anthropic, Gemini, and anything LiteLLM supports, with MCP integration, typed conversation history, and context providers for prompt construction. The monorepo (github.com/brainblend-ai/atomic-agents, note the rename from Eigenwise) adds a tool forge of copy-into-your-project tools, a TUI installer, and runnable examples. It has gathered roughly 6,000 GitHub stars and installs via pip as atomic-agents.
What it can do
- Schema-driven agents with typed input/output contracts\nBuilt on Instructor + Pydantic with LiteLLM provider support\nComposable atomic components plus a community tool forge\nTyped chat history, MCP, and multimodal content support
Setup
pip install atomic-agents (uv recommended); see the monorepo README
Prerequisites
Python 3.12+, an API key for your chosen LLM provider
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 101; Trending score 74/100 from researcher assessment.
★ 6,058 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.