Launch Builder
Flow System

Workflows As Ordered Node Graphs

A flow is the core execution unit in IAI. It is a set of nodes connected in a directed graph that the runtime can validate, execute, retry, and observe.

What Is A Flow

A flow is a workflow composed of ordered actions. The platform stores a graph definition, validates it, then turns it into execution steps.

Practical rule

If it cannot be serialized as nodes and edges, it is not yet a first-class flow artifact.

Flow Structure

{
  "nodes": [],
  "edges": []
}

The shape stays intentionally small. Complexity belongs inside node configuration, not in ad hoc flow-level nesting.

Node Types

input

Introduces data or user-provided values into the workflow.

condition

Evaluates a branch and chooses execution direction.

transform

Changes data shape or computes a derived payload.

output

Produces the terminal result of a flow.

http

Calls external HTTP services or internal APIs.

ai (future)

Reserved for model prompts, inference actions, or agentic steps.

Execution

Create Execution

The API or UI creates a run record linked to a flow definition.

Run

The runtime maps graph state into ordered node execution and stores logs.

Store Result

The final output and side-effects become part of execution history.

{
  "flowId": "flow_123",
  "status": "completed",
  "steps": [
    { "nodeId": "input-1", "status": "ok" },
    { "nodeId": "transform-1", "status": "ok" },
    { "nodeId": "output-1", "status": "ok" }
  ],
  "result": {
    "message": "Flow completed"
  }
}

UI Flow Builder

flow.iai.one is the current builder surface.

Drag & Drop

Compose nodes quickly without writing flow JSON by hand.

Preview

Inspect node configuration and flow structure before execution.

Run

Execute a draft and inspect runtime output with low friction.

Annotated Walkthrough

Annotated walkthrough for flow.iai.one showing node palette, canvas graph, run control, and execution output.
Guide frame for the current Flow model: palette on the left, node graph in the middle, and execution feedback on the right.
Step 1

Start from the node palette and add the smallest possible graph: one input, one logic or transform node, and one output.

Step 2

Connect nodes left to right so execution order is obvious before you worry about advanced branching.

Step 3

Run once and inspect execution state immediately. A first success means validation passed, steps executed, and result was stored.

Step 4

If a side-effect matters, chain the result into mail or a downstream system only after the base graph works reliably.

Flow To Mail Integration

One of the first practical use cases is sending flow outputs into the communication layer.

{
  "action": "mail.send",
  "to": "operator@example.com",
  "subject": "Flow completed",
  "body": "Execution output is ready."
}
System pattern

Flows compute. Mail delivers. App surfaces review. Keeping those responsibilities separate prevents product boundaries from collapsing into one giant surface.