Build agents
Edges
Connecting nodes and defining conversational paths
Edges connect nodes, defining the possible paths a conversation can take through your Agent Graph. They determine the flow logic and create structured conversational experiences.
Functionality
- Connections: Edges direct flow from a source node to a target node.
- Directionality: Each edge has a specific direction, flowing from one node to another.
- Visual Representation: Displayed as arrows on the Canvas.
Conditional Edges
A powerful feature of edges is the ability to add conditions, allowing for dynamic conversation paths based on context:
- Natural Language Conditions: Write conditions in plain English that the agent will interpret.
- Context Awareness: Conditions are evaluated based on the current conversation state, including user inputs and any variables.
- Multiple Paths: When a node has multiple outgoing edges, the agent evaluates the condition associated with each edge after executing the source node’s logic.
- Path Selection: The agent will follow the outgoing edge whose condition evaluates to true.
Examples
Example 1: Simple Conditional Routing
Example 2: Intent Detection
Example 3: Entity-Based Routing
Example 4: Complex Logic
Best Practices
- Clear Conditions: Make conditions clear and distinct to avoid ambiguity
- Condition Testing: Test conditions thoroughly with various user inputs
- Simplicity: Keep conditions focused on a single concept when possible
Building Complex Flows
By combining nodes and conditional edges, you can build sophisticated conversation flows that adapt to user inputs and context. Consider these patterns:
- Decision Trees: Use conditional edges to create branching paths based on user choices
- Information Collection: Route through different nodes to gather specific information
- Error Handling: Create paths for handling misunderstandings or invalid inputs
- Context Switching: Allow users to change topics with edges that detect new intents