Training: Creating Your First Workflow

This tutorial will walk you through building, testing, and running your first AI-powered automation.

🏗️ Phase 1: Designing the Flow

* Drag a "Manual Trigger" from the sidebar onto the canvas.

Drag an "AI Agent" node onto the canvas. Connect them: Click the small circle on the right of the Trigger and drag a line to the circle on the left of the AI Agent.

Click the AI Agent node to open the Configuration Panel on the right. Enter a System Prompt: "You are an assistant that translates any input into French." * Select one of the AI Models enabled for your workspace.

Use the description field in the editor to explain what the workflow does. This makes large workflow lists easier to scan later.

1. Open the Editor: From the Workflows page, click "Create Workflow".
2. Add a Trigger: Every flow must start with a Trigger node.
3. Add an AI Agent:
4. Configure the Agent:
5. Add a Description (recommended):

🧪 Phase 2: Testing & Variables

Workflows are powerful because they pass data between nodes.

Update Prompt: "Translate this text to French: {{input}}"*

Enter "Hello World" in the test payload box. Watch the nodes light up as they execute!

For Tool nodes, use Verify Setup to confirm connections/credentials are ready. Use Dry Run for safe read/list/check actions before moving to live runs.

1. Dynamic Input: In your AI Agent's prompt, you can use {{input}} to refer to data from the previous node.
2. Test Run: Click the "Run" button in the top toolbar.
3. Verify Tools Before Running:

🚀 Phase 3: Deployment & Monitoring

Once you are happy with your workflow:

1. Save: Click the Save icon to commit your changes.
2. Check Preflight: Open Validation and review the Preflight summary. It tells you whether the workflow is ready for test mode, publishing, and live traffic.
3. Choose the right environment: New workflows start in Development. Development and Staging are safe for testing, but only Production workflows can receive live traffic.
4. Publish when ready: New workflows start as Drafts. Drafts are safe for editor testing, but they do not receive live traffic.
5. Promote carefully: A strong path is Development → Staging → Production. Promote only after Preflight is clear and your dry runs look good.
6. Manual Execution: The editor can start test runs for drafts, paused workflows, and non-production environments. Live manual runs require the workflow to be active, published, and in Production.
7. Automated Execution: Schedule, email, webhook, and event triggers run live only after the workflow is active, Published, and in Production.
8. Monitor: If something goes wrong, check the Executions tab. Red nodes indicate where the flow stopped, and the error message will tell you why (e.g., "Invalid API Key" or "Rate Limit Exceeded").
9. Recover safely: If a job gets stuck in queue or a worker dies mid-run, NeuronFlow now flags the execution for recovery so you can requeue it without duplicating already-completed work.
10. Read the analytics: The Executions page now highlights success rate, average duration, retried runs, queue attention items, daily activity, and top failure reasons so you can improve reliability over time.

🧪 Note on AI‑Generated Workflows

Workflows created with the AI generator start paused by default to protect usage and avoid accidental auto‑runs.

• You can still run them in the editor — this uses test mode.
• To enable live/scheduled runs, activate the workflow from /workflows/.
• When the generated flow still needs setup, the editor shows an AI Workflow Review banner and opens the Validation panel so you can fix missing inputs or connections quickly.

💡 Pro Tips for Success

• Handle Errors: Use the Switch node to check if an API call was successful before proceeding to the next step.
• Use Retries Carefully: Add retries for transient failures, but set an Idempotency Key for write actions like create/update/delete/send.
• Build email auto-replies the easy way: Use Email Trigger → AI Agent → Gmail/SMTP Email. Set the recipient to {{trigger.from_email}}, keep the subject as Re: {{trigger.subject}}, and use {{input}} for the reply body.
• Use Memory v2: Store cursors or checkpoints with the Memory node (set/append) and keep sessions isolated with a Session ID.
• Use Context Memory: Save rules or brand voice once and auto-inject them into agent prompts across runs.
• Design inside limits: If you hit max nodes/edges/branches/tools-per-agent, split your flow into smaller workflows and chain them with webhooks or schedules.
• Loop responsibly: Large arrays and deep nesting are capped by plan runtime protections. Use batching (split arrays) for large workloads.
• Use Versions Before Big Changes: Save before risky edits. If a change makes the workflow worse, restore a previous version from the editor.
• Use environments intentionally: Build in Development, validate in Staging, and keep Production as the only place that receives real traffic.
• Audit tools regularly: The Admin → Tools page now includes a reliability audit so you can spot missing contracts or broken tool classes before customers hit them.
• Draft with AI first: Before building manually, try Architect with AI on the dashboard to get a suggested starting graph that you can review, validate, and edit.