Agent Orchestrator Workflow

in a very short paragraph in my style explain this: within each agent you can define functions and they are callable by the orchestrator connected on a graph controlling the agents so it is able to delegate tasks to the agents or submodules which act as a set of tools and memory that each can work on a task and complete it returned in a dicyclic graph so asyncronously it completes

What I built is essentially a graph of specialized agents where each node exposes callable functions that the orchestrator can invoke as needed. The orchestrator traverses the graph, delegates work to agents or submodules that contain their own tools and memory, and then asynchronously collects the results as tasks complete. Instead of a single monolithic workflow, computation flows through a directed graph, allowing independent agents to solve parts of a problem in parallel and return their outputs back into the execution graph for further processing.