Workflow Configuration

Workflows

Dryade workflows orchestrate multiple AI agents to complete complex tasks. Workflows define the execution order, data flow, and escalation rules between agents.

Supported Frameworks

• CrewAI: Role-based agent crews with task delegation
• LangGraph: Graph-based agent workflows with state management
• Google ADK: Agent Development Kit for structured agent pipelines

Workflow Definition

Workflows can be defined as markdown files or full code pipelines:

name: customer-support-triage
description: Classify and route support tickets
agents:
  - name: classifier
    framework: crewai
    task: Classify the incoming ticket by urgency and category
  - name: responder
    framework: langraph
    task: Draft an initial response based on classification
steps:
  - agent: classifier
    input: "{{ ticket.content }}"
    output: classification
  - agent: responder
    input: "{{ classification }}"
    condition: "classification.urgency != 'critical'"

Visual Workflow Editor

The Dryade Workbench includes a visual workflow editor for designing agent pipelines without writing YAML. Drag-and-drop agents, connect data flows, and configure escalation rules.

Execution Modes

• Sequential: Agents execute one after another
• Parallel: Independent agents run simultaneously
• Conditional: Branching based on agent outputs
• Loop: Repeat steps until a condition is met

Escalation Rules

Configure when workflows should escalate to human operators:

• Confidence thresholds
• Error count limits
• Timeout-based escalation
• Content-based triggers

Monitoring

Every workflow execution is logged with:

• Step-by-step execution trace
• Agent input/output at each step
• Token usage and cost tracking
• Duration and performance metrics