Repolode
Submit
Eigenwise

Eigenwise/atomic-agents

A lightweight Python framework for building modular, schema-driven AI agent pipelines on top of Instructor and Pydantic.

View on GitHubWebsite
Stars
6,320
Forks
562
Open issues
12
Last commit
9 days ago

What is it?

Atomic Agents is a Python framework for building LLM-based agents and pipelines out of small, single-purpose components. Each agent is defined by a system prompt, an input schema, an output schema, a chat history, and optional context providers that inject dynamic information into the prompt at runtime. Schemas are Pydantic models, so inputs and outputs are validated and predictable.

Because an agent's output schema can match another tool's input schema, agents and tools can be chained together or swapped without rewriting the surrounding code. All control flow is plain Python. The framework is built on Instructor, which means any provider Instructor supports, such as OpenAI, Anthropic, Groq, Gemini, or Ollama, can be used.

The repository also includes Atomic Forge, a set of ready-made tools such as web search, Wikipedia, YouTube transcript scraping, and PDF reading, along with an interactive CLI called Atomic Assembler for copying them into a project. Example applications cover RAG chat, deep research, and orchestration agents.

Key features

  • 01Agents are defined by system prompt, Pydantic input and output schemas, and chat history.
  • 02Context providers inject dynamic information into the system prompt at runtime.
  • 03Agents and tools chain together by aligning output and input schemas.
  • 04Works with any provider supported by Instructor, including OpenAI, Anthropic, Groq, Gemini, and Ollama.
  • 05Atomic Forge offers prebuilt tools for search, scraping, PDF reading, and more.
  • 06Each run is emitted as an OpenTelemetry span for observability.

Use cases

  • Building structured-output chatbots or extraction pipelines where predictable schemas matter.
  • Prototyping RAG or web-search agents from reusable tool components.
  • Swapping LLM providers or search backends without rewriting application logic.
  • Adding tracing to LLM applications through any OpenTelemetry-compatible backend.

Quick start

  1. 1

    Install the framework from PyPI

    pip install atomic-agents
  2. 2

    Install a provider SDK if you need one other than OpenAI (e.g. Anthropic)

    pip install instructor[anthropic]
  3. 3

    Write an agent using the Quick Example from the README and run it with Python

Things to know

  • Requires an API key from a model provider such as OpenAI, Groq, Anthropic or Gemini to run agents.

In collections

Open-source AI agent frameworks & apps

Topics

#ai#llm#agents#python#pydantic#openai#structured-output#rag#artificial-intelligence#large-language-model

FAQ

What is atomic-agents?
A lightweight Python framework for building modular, schema-driven AI agent pipelines on top of Instructor and Pydantic. Atomic Agents is a Python framework for building LLM-based agents and pipelines out of small, single-purpose components. Each agent is defined by a system prompt, an input schema, an output schema, a chat history, and optional context providers that inject dynamic information into the prompt at runtime. Schemas are Pydantic models, so inputs and outputs are validated and predictable.
Is atomic-agents free to use? What license does it have?
atomic-agents is open source under the MIT license. Check the license terms before using it in commercial products.
Is atomic-agents still maintained?
Actively maintained. The last commit to atomic-agents was on Oct 29, 2026.
What can I use atomic-agents for?
Building structured-output chatbots or extraction pipelines where predictable schemas matter. Prototyping RAG or web-search agents from reusable tool components. Swapping LLM providers or search backends without rewriting application logic. Adding tracing to LLM applications through any OpenTelemetry-compatible backend.
How do I install atomic-agents?
Install the framework from PyPI: pip install atomic-agents Install a provider SDK if you need one other than OpenAI (e.g. Anthropic): pip install instructor[anthropic] Write an agent using the Quick Example from the README and run it with Python
Where can I get atomic-agents?
The source code and installation instructions are on GitHub: https://github.com/Eigenwise/atomic-agents

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects