Building Automations
20 steps - 10 minutes
Learn how to build AI-powered automations to streamline your workflows.
What are Automations?
Automations are graph-based AI workflows built with ReactFlow. They consist of nodes (trigger, agent, decision, guardrail, human_approval) and edges (connections). Agent nodes use data.config.persona_id, prompt_id, context_ids, and tools (namespaces/subtools). Decision nodes provide if-then-else branching with simple conditions (Basic mode) or AI-powered routing (Agentic mode). Execution uses SSE streaming with durable transcript snapshots for history hydration. Test runs execute using the current in-builder (unsaved) state.
Navigate to Automations
From the sidebar, click on "Automations" to access the automation builder.
Click "Create Automation"
Click the "Create Automation" button to start building a new workflow.
Enter Automation Name and Description
Give your automation a descriptive name (e.g., "Daily Email Summary", "Lead Qualification") and add a description explaining what it does.
Understanding the Visual Builder
The builder uses ReactFlow for the graph-based workflow. You can drag nodes, connect them with edges, and configure each node. The graph structure includes nodes (trigger, agent, decision, guardrail, human_approval) and edges (connections).
Choose a Trigger Type
Select how your automation starts: Schedule (cron-based), Webhook (external trigger), App Event (platform event), or Manual (on-demand).
Configure Trigger Settings
For scheduled automations, set the cron expression. For webhooks, copy the webhook URL. For app events, select the event type.
Add an Agent Node
Drag an Agent node onto the canvas from the node palette. This is where the AI will perform tasks using agents, prompts, context, and tools.
Select an Agent for the Automation
Drag an agent from the side panel into the agent node. The agent avatar and name will fill the node prominently. This sets the agent runtime identity.
Add a Prompt to the Agent
Drag a prompt into the agent node to define what task the AI should perform. This sets data.config.prompt_id and loads the full prompt into the agent.
Add Context Blocks (Optional)
Drag context blocks into the agent node to provide relevant background information. This sets data.config.context_ids as an array.
Whitelist Tools - Namespaces
Drag namespace cards from the side panel into the Tools section. Namespaces show app images initially and expand to show subtools when clicked.
Whitelist Tools - Specific Subtools
When a subtool is selected in the builder, its full schema is loaded directly into the LLM prompt. Tools are executed via namespace wrappers with tool_name parameter.
Configure Node Outputs
In the Outputs section (separate from Tools), define what data this node should output for use in subsequent nodes.
Connect Nodes with Edges
Drag from the output handle of one node to the input handle of another to create edges. This defines the workflow execution order.
Add More Nodes (Optional)
Add additional agent nodes, decision nodes (for conditional branching), guardrail nodes, or human approval nodes to create multi-step workflows. Use outputs from previous nodes as inputs.
Test Run with Current State
Click "Test Run" to execute your automation using the current in-builder (unsaved) state. Watch the execution in real-time via SSE streaming.
Review Test Results via SSE Events
Execution streams live events over SSE and materializes durable transcript history for fast run replay.
Save Your Automation
Click "Save" to store your automation. You will see a confirmation prompt on reload/leave if there are unsaved changes.
Enable Automation
Toggle the automation to "Active" to enable scheduled runs or webhook triggers.
Monitor Execution History
View execution history to track when your automation runs, review results, and see detailed event timelines.
Best Practices
- •Start simple and add complexity gradually
- •Test thoroughly before enabling scheduled runs
- •Use clear, specific prompts for each agent node
- •Whitelist only the tools needed for each task
- •Monitor execution history to identify and fix issues
- •Use Uru Architect to help design complex workflows
Next Step
Learn how to use Uru Architect to build automations with natural language.
Using Uru Architect