GraphAI is a minimalistic “AI framework” that aims to not be an AI framework at all. Instead, it provides a simple, flexible graph-based architecture that engineers can use to develop their own AI frameworks and projects.
GraphAI is a lightweight library built around the concept of a computational graph. It provides:
Unlike other AI libraries, GraphAI doesn’t ship with predefined concepts of “LLMs”, “Agents”, or other high-level AI abstractions. Instead, it gives you the tools to build these concepts yourself, exactly how you want them.
Many existing AI frameworks impose their view of what AI applications should look like, creating a “local minimum” that constrains innovation. GraphAI takes a different approach:
AI applications frequently rely on API calls that involve significant waiting time. GraphAI is built from the ground up to be async-first, allowing your Python code to efficiently handle these operations rather than wasting compute cycles while waiting for responses.
At its core, GraphAI provides a graph of connected nodes where:
This architecture makes complex workflows simple to understand and modify.
GraphAI provides just what you need, nothing more:
Graph
class for orchestrating executionNode
and Router
decorators for defining processing unitsCallback
for streaming and communicationState
management for maintaining contextConsider GraphAI when: