Agentic Workflows Overview
You already have an application with data and an agent that can reason about it. An Agentic Workflow turns a process you repeat by hand into a graph of steps that runs the same way every time. Examples include a support ticket needing consistent triage, a report generated on the same schedule, and an order needing the same checks before it ships.
You author a workflow by describing the process to the C3 Code agent, which builds and publishes it. To build a workflow in code instead, use the Workflows Python API from a Jupyter notebook.
What a workflow is made of
A workflow is a graph of connected steps that the platform runs on demand or on a schedule. Four pieces make up every workflow:
- Nodes: the individual steps. A node calls a large language model (LLM), asks a person for input, runs code, or calls a tool. Each node starts from a node template, a reusable blueprint for one type of step.
- Edges: the connections between nodes that decide what runs next. A plain edge always continues to the same next node, while a conditional edge branches to a different node depending on a value in the state.
- State: the shared data a workflow carries as it runs. Every node reads its inputs from the state and writes its outputs back to the state. A value one node produces is available to every node after it.
- Interrupts: a pause a workflow takes when a node needs a person to answer something before it continues. A node might ask someone to approve a result, or supply a value the workflow still needs.
Author a workflow in C3 Code
Currently, the creation and authoring experience for workflows is only available through the C3 Code agent. To learn how to create a workflow in your C3 Code Application, see Author a Workflow with the Agent.
List and run workflows in C3 AI Studio
The current features allowed via the C3 AI Studio UI are to view the list of your published workflows, click a workflow and run it. To learn more, see List and Run Workflows via C3 AI Studio.
Configure workflows in Jupyter
Build a workflow in code when you want to register each node template, create each node, and connect the edges yourself. Open Jupyter Notebook from your application in C3 AI Studio, set the py-workflows runtime as the notebook kernel, and use the Workflows Python API to create, validate, publish, and run the workflow.
See Configure Workflows through Jupyter.