Workflow Pipeline

Build an automated classify-then-respond pipeline using the workflow builder. Time: ~20 minutes.

What You Will Build

A workflow that classifies incoming messages by intent
Conditional routing to different response agents based on classification
An automated pipeline: Input → Classify → Route → Respond
Error handling and fallback paths

What is a Workflow?

A workflow is a directed acyclic graph (DAG) of nodes that process data in sequence or in parallel. Each node can be an agent call, a condition check, a transformation, or an API call. Workflows let you build complex automation without writing code.

Step-by-Step Guide

Create Helper Agents

First, create the agents that will be used as nodes in the workflow.

# Classifier Agent - determines intent
curl -X POST http://localhost:8000/api/v1/agents \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "intent-classifier",
    "description": "Classifies user messages into categories",
    "model": "gpt-4o-mini",
    "system_prompt": "Classify the user message into exactly one category: billing, technical, general, or complaint. Respond with only the category name, nothing else.",
    "tags": ["workflow", "classifier"]
  }'

# Response agents for each category
curl -X POST http://localhost:8000/api/v1/agents \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "billing-responder",
    "model": "gpt-4o-mini",
    "system_prompt": "You handle billing inquiries. Be helpful with payment, invoice, and subscription questions."
  }'

curl -X POST http://localhost:8000/api/v1/agents \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "technical-responder",
    "model": "gpt-4o",
    "system_prompt": "You handle technical support. Provide detailed troubleshooting steps."
  }'

curl -X POST http://localhost:8000/api/v1/agents \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "general-responder",
    "model": "gpt-4o-mini",
    "system_prompt": "You handle general questions. Be friendly and informative."
  }'

Create the Workflow

Define the workflow structure with nodes and edges.

curl -X POST http://localhost:8000/api/v1/workflows \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "support-pipeline",
    "description": "Classify incoming messages and route to the right responder",
    "nodes": [
      {
        "id": "input",
        "type": "input",
        "config": { "description": "Incoming user message" }
      },
      {
        "id": "classify",
        "type": "agent",
        "config": { "agent_id": "CLASSIFIER_AGENT_ID" }
      },
      {
        "id": "router",
        "type": "condition",
        "config": {
          "conditions": [
            { "field": "output", "operator": "contains", "value": "billing", "target": "billing" },
            { "field": "output", "operator": "contains", "value": "technical", "target": "technical" },
            { "field": "output", "operator": "contains", "value": "complaint", "target": "escalate" }
          ],
          "default_target": "general"
        }
      },
      {
        "id": "billing",
        "type": "agent",
        "config": { "agent_id": "BILLING_AGENT_ID" }
      },
      {
        "id": "technical",
        "type": "agent",
        "config": { "agent_id": "TECHNICAL_AGENT_ID" }
      },
      {
        "id": "general",
        "type": "agent",
        "config": { "agent_id": "GENERAL_AGENT_ID" }
      },
      {
        "id": "escalate",
        "type": "webhook",
        "config": {
          "url": "https://your-ticketing-system.com/api/tickets",
          "method": "POST"
        }
      },
      {
        "id": "output",
        "type": "output",
        "config": { "description": "Final response to user" }
      }
    ],
    "edges": [
      { "from": "input", "to": "classify" },
      { "from": "classify", "to": "router" },
      { "from": "billing", "to": "output" },
      { "from": "technical", "to": "output" },
      { "from": "general", "to": "output" },
      { "from": "escalate", "to": "output" }
    ]
  }'

Validate the Workflow

Check that the workflow DAG is valid before running.

curl http://localhost:8000/api/v1/workflows/WORKFLOW_ID/validate \
  -H "Authorization: Bearer YOUR_API_KEY"

# Response:
# { "valid": true, "node_count": 8, "edge_count": 6 }

Execute the Workflow

Run the workflow with a test input.

# Test with a billing question
curl -X POST http://localhost:8000/api/v1/workflows/WORKFLOW_ID/execute \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "input": "I was charged twice on my last invoice" }'

# Test with a technical question
curl -X POST http://localhost:8000/api/v1/workflows/WORKFLOW_ID/execute \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "input": "My API calls are returning 500 errors" }'

View Execution History

See how the workflow routed each request.

curl http://localhost:8000/api/v1/workflows/WORKFLOW_ID/executions \
  -H "Authorization: Bearer YOUR_API_KEY"

# Response shows each execution with:
# - Input message
# - Classification result
# - Which path was taken
# - Final response
# - Total execution time

Pipeline Visualization

Workflow DAGtext

[Input] --> [Classifier Agent]
                |
                +-- "billing"   --> [Billing Responder]   --> [Output]
                |
                +-- "technical" --> [Technical Responder]  --> [Output]
                |
                +-- "complaint" --> [Webhook: Escalate]    --> [Output]
                |
                +-- (default)   --> [General Responder]    --> [Output]

Using the Dashboard

The Workflows tab provides a visual drag-and-drop builder:

Navigate to the Workflows tab
Click Create Workflow
Drag nodes from the palette: Input, Agent, Condition, Webhook, Output
Connect nodes by dragging from output ports to input ports
Configure each node by clicking on it (select agent, set conditions, etc.)
Click Validate to check the DAG
Click Execute to run with test input

Advanced Patterns

Use parallel nodes to run multiple agents simultaneously
Add transform nodes to reshape data between steps
Set timeout on each node to prevent hanging
Use retry configuration for unreliable external APIs

Embed Widget

Cost Control