Workflow Builder Guide

The workflow builder is where you design, test, publish, and operate NeuronFlow automations. Use this guide when you need to understand what each builder state means before a workflow receives live traffic.

Builder Basics

Open the builder from /app/ or from a workflow card. A workflow is made from:

• one Trigger or Webhook start node
• action, AI, logic, approval, memory, and tool nodes
• edges that describe how data moves from one node to the next
• an optional Output node for the final response shown in execution history

You can save a draft while it is still incomplete. Running a workflow is stricter: the graph must have a valid start node and no blocking validation errors.

Validation And Preflight

The builder validates the graph while you work. It blocks unsafe or impossible flows, including:

• more than one Trigger or Webhook
• cycles that could create infinite loops
• edges into trigger nodes
• invalid Tool-to-Agent wiring
• plan limit violations for nodes, edges, branches, or tools attached to one agent

The Preflight view explains whether the workflow is ready for:

• editor test runs
• publishing
• live traffic
• environment promotion
• node setup, credentials, and required tool inputs

Warnings do not always block saving, but they should be reviewed before production use.

Saving, Testing, And Dry Runs

Use Save to keep the latest graph, lifecycle state, environment, and Agent Mode setting.

Use Run in the editor to start a test or live execution:

• drafts run in test mode
• Development and Staging workflows run in test mode
• paused workflows can still be tested from the editor in test mode
• active, published Production workflows can run in live mode

For Tool nodes, use Verify Setup to confirm the selected connection or inline credentials are available. Use Dry Run only for safe read/list/check actions. Destructive actions such as create, update, delete, send, or charge should be tested with intentionally safe inputs.

Lifecycle States

NeuronFlow separates editing safety from live traffic.

Draft

Draft workflows can be edited and tested in the builder. They do not receive live schedule, email, webhook, or event traffic.

Published

Published workflows are eligible for live traffic only when they are also in the Production environment and activated.

Environments

Use environments to keep production behavior intentional.

• Development: build and iterate safely
• Staging: validate the final shape before live use
• Production: the only environment that can receive real live traffic

Manual editor runs are still useful outside Production because they run as test executions.

Activation

The activation toggle controls live traffic.

• Active means a published Production workflow can receive live manual runs, schedules, webhooks, email triggers, and event traffic.
• Paused means live traffic is blocked. The builder can still start a test run so you can debug without reactivating the workflow.

The Workflows list only enables Run Now for workflows that are active, published, and in Production.

Approvals And Revision Loops

Approval nodes pause execution until a reviewer chooses an explicit action:

• Approve continues through the approved branch
• Reject follows the rejected branch
• Request Changes follows the changes-requested branch when that action is enabled

Request-changes loops are meant for correction iterations. For example, a reviewer can ask for better wording, the workflow can route the feedback back to the previous work step, and the revised result can return to approval. Set a revision limit so the workflow cannot loop forever.

Executions waiting for approval should be continued through the approval action, not through a generic resume action.

Versions And Restore

Every save, generation, and restore creates a workflow version. Use versions before risky edits so you can return to a previous graph if a change breaks the workflow.

Restoring a version also restores the workflow graph and saved node secret references captured with that version.

Simple Mode Vs Agent Mode

Standard workflow mode is best for short, deterministic flows where the node graph fully describes the work.

Agent Mode is best when a workflow needs memory, checkpoints, recovery across runs, or a readable plan. The Agent Mode plan explains intent, but the node graph still controls execution.

Reading Results

After a run starts, use Executions or the builder history panel to inspect:

• execution status
• node-by-node input and output
• approval decisions and comments
• retries, duration, and queue/recovery events
• final Output data

This is the fastest way to confirm that the workflow is behaving as designed before you activate live traffic.