Skip to main content

Overview

Agent Graph
Agent Graphs are the heart of every Assistant. They define logic, decisions, and outcomes through nodes on a visual canvas. Agent Graphs are designed and maintained in Graph Studio. Assistants execute Agent Graphs to deliver outcomes.

Key concepts

  • Nodes: Start, Master Agent, Child Agent, A2A Agent, Tool, End
  • RAG and Tools within Agent nodes
  • Variables and Conditional Edges

Anatomy of a Agent Graph

  • Start Node — entry point connected to a Trigger or Intent.
  • Agent Nodes — actions or messages the AI agent performs.
  • Tool nodes — invoke custom tools or APIs.

Using Aura in Graph Studio

  • Describe the goal in natural language.
  • Aura generates a full graph layout with connections and node suggestions.
  • You can review and modify before saving.

Debugging Agent Graphs

  • Always validate input/output connections before saving.
  • Use the Preview Run to simulate execution.
  • Check the Execution Logs in the environment for variable mismatches or failed tool calls.

Best practices

  • Keep flows modular
  • Write the quality prompts
  • Reuse Child Agents for common patterns
  • Use variables consistently for data passing