Skip to content

Usage overview

  1. Open a workflow in the FlowDrop editor.
  2. Add AI Agent Executor from the Agents category.
  3. Pick the agent in the AI Agent field. The selector searches agents by label and previews the selected agent's description and tools. It also links to the form for creating a new agent.
  4. Connect an upstream node to the task_description input. This input has no form field; it always comes from another node, for example a text input or a chat input node.
  5. Connect the outputs you need, usually response, success and structured_output.
  6. Run the workflow.

What happens on a run

flowchart TD
    Start([Node runs]) --> Validate{Parameters valid?}
    Validate -- no --> VFail[Validation error, node does not run]
    Validate -- yes --> Build[Build task and chat input]
    Build --> Overrides[Apply llm_model and tool_usage_limits]
    Overrides --> Solv{determineSolvability}
    Solv -- Solvable --> Solve[agent solves the task] --> Out1[success: true, structured_output]
    Solv -- Informs --> Out2[success: true, information]
    Solv -- Needs answers --> Pause[Workflow pauses with an interrupt]
    Pause -- person replies --> Build
    Solv -- Not solvable --> Out3[success: false, error_message]
    Build -. exception .-> Err[Job fails, error port fires]
    Solve -. exception .-> Err

Pages in this section