Workflows

Workflows let you build complex data pipelines visually by connecting nodes on a canvas. Chain together database queries, AI calls, code execution, and data transformations.

Creating a Workflow

1. Navigate to Workflows from the sidebar menu
2. Click New Workflow
3. Give your workflow a name

The Workflow Canvas

The canvas is a drag-and-drop interface for building pipelines:

• Nodes - Individual processing steps
• Edges - Connections between nodes showing data flow
• Mini-map - Overview for navigating large workflows

Navigation

• Pan - Click and drag the canvas
• Zoom - Scroll wheel or pinch gesture
• Select - Click a node to select it
• Multi-select - Shift+click or drag a selection box

Node Types

Database Query Node

Execute SQL queries against your connected data sources.

Configuration:

• Select a data source
• Write your SQL query
• Define input parameters
• Output: Query results as a table

Python Code Node

Run Python code in an isolated environment.

Configuration:

• Write Python code in the Monaco editor
• Access inputs from previous nodes
• Use pandas, numpy, matplotlib
• Output: Code execution results

Template Node

Process text templates with Jinja2.

Configuration:

• Write your template
• Map inputs to template variables
• Output: Rendered text

AI Model Node

Send prompts to an AI model and get responses.

Configuration:

• Select a model
• Configure temperature and other settings
• Define the prompt (can include outputs from previous nodes)
• Output: AI response

Connecting Nodes

1. Hover over a node to see connection handles
2. Drag from an output handle to an input handle
3. The connection shows data will flow between nodes

Data Flow

• Data flows from left to right (or top to bottom)
• Each node receives inputs from connected upstream nodes
• Outputs are passed to connected downstream nodes

Properties Panel

Click a node to open its properties in the side panel:

• Node settings - Configure the node’s behavior
• Code editor - Edit SQL or Python code
• Input mapping - Connect inputs from other nodes
• Output preview - See sample output data

Running Workflows

Execute All

Click Run to execute the entire workflow from start to finish.

Execute Node

Right-click a node and select Execute to run just that node (and its dependencies).

Execution Order

Nodes execute based on their connections:

1. Nodes with no inputs run first
2. Then nodes whose inputs are satisfied
3. Continues until all nodes complete

Saving and Loading

• Workflows auto-save as you work
• Use Export to download a workflow definition
• Use Import to load a saved workflow

Best Practices

• Start simple - Build and test small pieces first
• Name nodes clearly - Use descriptive names for easy understanding
• Test incrementally - Execute nodes individually as you build
• Handle errors - Add error handling in code nodes
• Document - Add comments in code nodes explaining logic

Example: Sales Report Workflow

A typical workflow might look like:

[Database Query] → [Python Transform] → [AI Summary] → [Template Output]
     ↓                    ↓                  ↓              ↓
  Fetch sales        Calculate KPIs    Generate insights  Format report

1. Database Query - Fetch raw sales data
2. Python Transform - Calculate metrics and trends
3. AI Summary - Have AI analyze the data
4. Template Output - Format everything into a report

Troubleshooting

Node Won’t Execute

• Check all required inputs are connected
• Verify upstream nodes completed successfully
• Check for errors in node configuration

Data Not Passing Through

• Verify connection handles are properly linked
• Check data types match between nodes
• Look at node output preview for issues

Code Errors

• Check Python syntax
• Verify imported libraries are available
• Look at error messages in execution results