Execution order in multi-branch workflows

When you design a workflow, it’s important to understand how Agentic WorkFlow decides the order in which nodes execute.
This ensures your data moves correctly from one step to the next, especially in workflows with multiple branches.

Execution overview

Agentic WorkFlow executes nodes using a breadth-first search (BFS) order.
That means all nodes at the same level (connected to the same upstream node) are executed before moving deeper into the workflow chain.

flowchart TB
    B["Extract Data"] L_B_D_0@--> D["Transform Data"]
    C["Load Config"] L_C_D_0@--> D
    D L_D_E_0@--> E["Send Email"]
    n1[" "] L_n1_B_0@--> B & C

    B@{ shape: rounded}
    D@{ shape: rounded}
    C@{ shape: rounded}
    E@{ shape: rounded}
    n1@{ icon: "fa:circle-play", pos: "b"}
    style B fill:#E8F5E9
    style D fill:#E1F5FE
    style C fill:#FFF3E0
    style E fill:#E8EAF6

    L_B_D_0@{ animation: slow } 
    L_C_D_0@{ animation: slow } 
    L_D_E_0@{ animation: slow } 
    L_n1_B_0@{ animation: slow } 
    L_n1_C_0@{ animation: slow }

How this works

The workflow starts from the trigger node.
Agentic WorkFlow identifies all directly connected nodes and executes them one by one.
Once all first-level nodes finish, it proceeds to their connected nodes.
The process repeats until all reachable nodes complete execution.

Execution priority

Each node waits for all its predecessors to finish before running. If one of the required nodes fails or is skipped, dependent nodes won’t execute until all prerequisites are completed.

This helps maintain data consistency across the workflow and prevents nodes from running with incomplete inputs.

Example: multi-branch workflow

flowchart TB
    n2["Extract user data"] L_n2_n4_0@--> n4["Combine results"]
    n3["Fetch order history"] L_n3_n4_0@--> n4
    n4 L_n4_n5_0@--> n5["Send report"]
    n6[" "] L_n6_n2_0@--> n2 & n3

    n2@{ shape: rounded}
    n4@{ shape: rounded}
    n3@{ shape: rounded}
    n5@{ shape: rounded}
    n6@{ icon: "fa:circle-play", pos: "b"}
    style n4 stroke:#00C853,fill:#C8E6C9

    L_n2_n4_0@{ animation: slow } 
    L_n3_n4_0@{ animation: slow } 
    L_n4_n5_0@{ animation: slow } 
    L_n6_n2_0@{ animation: slow } 
    L_n6_n3_0@{ animation: slow }

In this example:

“Extract user data” and “Fetch order history” will firstly be executed (the order depends of the creation order).
“Combine results” will wait for both branches to finish before running.
“Send report” will only execute after “Combine results” completes.

Example: complex multi-branch workflow

flowchart TB
    B["1"] L_B_B1_0@--> B1["4"] & n2["5"]
    D2["6"] L_D2_E_0@--> E["7"]
    E L_E_F_0@--> F["Continue Workflow"]
    n1[" "] L_n1_B_0@--> B & D["3"] & C["2"]
    B1 L_B1_E_0@--> E
    C L_C_E_0@--> E
    D L_D_D2_0@--> D2

    B@{ shape: rounded}
    B1@{ shape: rounded}
    n2@{ shape: rounded}
    D2@{ shape: rounded}
    E@{ shape: rounded}
    n1@{ icon: "fa:circle-play", pos: "b"}
    D@{ shape: rounded}
    C@{ shape: rounded}
    style E stroke:#00C853,fill:#C8E6C9

    L_B_B1_0@{ animation: slow } 
    L_B_n2_0@{ animation: slow } 
    L_D2_E_0@{ animation: slow } 
    L_E_F_0@{ animation: slow } 
    L_n1_B_0@{ animation: slow } 
    L_n1_D_0@{ animation: slow } 
    L_n1_C_0@{ animation: slow } 
    L_B1_E_0@{ animation: slow } 
    L_C_E_0@{ animation: slow } 
    L_D_D2_0@{ animation: slow }

Browser context considerations

When your workflow runs inside a browser environment, keep these factors in mind:

Page state changes: Some nodes interact with a webpage (clicks, scrolls) — ensure timing is handled properly.
Timing dependencies: Wait for page updates (e.g., after form submissions) before the next step.
Performance constraints: Browser-based workflows have limited resources; plan execution to avoid heavy parallel operations.