Build Your First AI-Generated Workflow in 5 Minutes
Describe your automation in plain English. The AI generates the workflow. Refine it visually. Deploy in seconds.
5-Step Automation Setup
Navigate to Blueprint
Open the Aeion OS admin panel and go to Blueprint > Workflows. You'll see the visual canvas, template library, and AI generator panel side-by-side. **Why it matters:** all three entry points sit next to each other, so you can start from a plain-English prompt, clone one of the pre-built templates, or drag nodes onto a blank canvas — whichever fits the automation in front of you.
Describe Your Automation
In the AI Generator panel, type your automation in plain English. Examples: "When a new contact is created in CRM, send a welcome email and add them to a Mailchimp list" or "Every Monday at 9 AM, generate a sales report and send it to the sales team." Select your AI provider (OpenAI, Anthropic, etc.) and model. **Why it matters:** the generator works from the live list of triggers and actions across your enabled modules, so it only proposes steps that actually exist in your OS — you won't get a workflow wired to a connector you don't have.
Review Generated Workflow
The AI generates a complete workflow with trigger, nodes, edges, and error handling. Check the confidence score (green = production-ready, yellow = review recommended). Review suggestions and warnings. Use the Refinement panel to: simplify, add error handling, add delays, add logging, or add notifications. **Why it matters:** the confidence score reflects how complete the generated workflow is, and each refinement command applies to the output of the last one — so you can harden a draft step by step and see a clear diff of exactly what changed at each pass.
Customize in Visual Editor
Drag-and-drop to reposition nodes. Click a node to edit its configuration (e.g., change email template, modify condition logic). Connect new nodes by dragging from output ports to input ports. The editor validates in real-time—red borders for errors, green for valid. **Why it matters:** validation runs as you edit, so a broken edge or a missing required field surfaces on the canvas before you ever deploy — not as a silent failure the first time a real trigger fires.
Test & Deploy
Use the "Test Run" button to execute with sample data. Check the execution trace for each node's input, output, and duration. Fix any failures. When satisfied, click "Activate" to deploy. The workflow starts listening for triggers immediately. **Why it matters:** the per-node execution trace shows the exact input, output, and duration of every step against your sample data, so you see precisely where a workflow would break before it ever touches a live trigger.
typescript
// Example: Generate a workflow via the AI generation APIconst result = await fetch("/api/v1/blueprints/ai/generate", { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ prompt: "When a new commerce order is placed, send a confirmation email and notify the fulfillment team in Slack", providerId: "openai", modelId: "gpt-4-turbo", }),}).then((r) => r.json());// result.data.confidence === 0.92 (high confidence)// result.data.workflow.nodes.length === 5
// Example: Apply a refinement commandconst refined = await fetch("/api/v1/blueprints/refine", {method: "POST",headers: { "Content-Type": "application/json" },body: JSON.stringify({workflow: result.data.workflow,command: "add_error_handling",}),}).then((r) => r.json());
// Example: List workflow generation history (max 50 per page)const history = await fetch("/api/v1/blueprints/history?limit=20&favorites=false",).then((r) => r.json());// history.data[0].confidence === 0.92
// Example: Replay a generation with a different modelconst replay = await fetch(`/api/v1/blueprints/history/${history.data[0].id}/replay`,{method: "POST",headers: { "Content-Type": "application/json" },body: JSON.stringify({ changeModel: "claude-3-5-sonnet-20241022" }),},).then((r) => r.json());The generator always works from the live list of triggers and actions across your Aeion OS modules—every module contributes its own actions automatically. The AI receives a comprehensive list including action names, descriptions, and configuration options.
Scores of 0.85+ are recommended for immediate production use. Scores between 0.6-0.84 benefit from reviewing suggestions and warnings. Scores below 0.6 should be heavily reviewed or regenerated with a more specific prompt.
Yes. If your provider is registered in the Aeion AI registry, it appears in the dropdown. If not, you can add it in Platform > AI Settings. The generator supports any provider that implements the AI adapter interface.
Apply them sequentially: `simplify` → `add_error_handling` → `add_logging`. Each refinement operates on the output of the previous, and a clear diff tracks what was added, modified, or removed at each step.
Yes. The 50+ built-in actions cover every module: CRM (contacts, deals, pipelines), Commerce (orders, payments, fulfillment), Helpdesk (tickets, routing, SLA), Connect (SMS, calls, broadcasts), Workspace (pages, databases, meetings), Finance (invoices, journal entries). Cross-module workflows are first-class citizens.
Blueprint ships 23 pre-built templates across categories (Sales, Marketing, HR, Finance, etc.). Browse by category, difficulty, or search by name/description. Click "Use Template" to copy it to your canvas. Templates include the full workflow definition, documentation, and recommended provider/model settings.
Event triggers (`type: "event"`) listen to Aeion OS events in real-time—e.g., `record.created`, `record.updated`, custom module events. Schedule triggers (`type: "schedule"`) use cron expressions for time-based execution—e.g., `"0 9 * * 1"` for Monday at 9 AM. Webhook triggers accept inbound HTTP POSTs. Manual triggers are one-click executions.
History is scoped per user within your organization. Each user keeps up to 50 records; once you'd exceed that, the oldest non-favorited record is deleted to make room for the new one. Favorited records are exempt from that cleanup — there's no separate age-based expiry.