Workflow Designer

The Workflow Designer lets you create, configure, and execute multi-agent AI workflows — visually or through natural language.

---

Overview

Synaptiq workflows coordinate multiple AI agents to solve complex problems. Each workflow consists of:

• Nodes — individual AI agents with specific roles and system prompts
• Edges — connections defining execution order and data flow
• Flow Type — how agents are orchestrated (sequential, parallel, supervisor, dynamic)
• Input/Output — data contracts for the workflow's inputs and results

---

Creating Workflows

Method 1: Natural Language Generation

Describe your workflow in plain English:

"Create a workflow for generating ABA therapy goals. 
 It should have a supervisor agent that coordinates 
 four specialists: ABA, Speech Therapy, OT, and CBT. 
 Each specialist generates domain-specific goals, 
 and the supervisor consolidates them."

Synaptiq's workflow generation engine:

1. Parses the description
2. Identifies required agents and their roles
3. Generates system prompts for each agent
4. Determines the optimal flow type
5. Creates the workflow definition

Method 2: Visual Designer

The drag-and-drop visual designer provides:

• Node palette — available agent types and tools
• Canvas — visual flow editor with grid snapping
• Properties panel — configure each node's prompt, model, and parameters
• Connection drawing — link nodes with edges to define execution flow

Method 3: API

POST /api/v1/workflow/save
Content-Type: application/json

{
  "spec": {
    "name": "Customer Onboarding",
    "description": "Multi-step customer onboarding workflow",
    "type": "SEQUENTIAL",
    "nodes": [
      {
        "id": "welcome",
        "name": "Welcome Agent",
        "systemPrompt": "You are a friendly onboarding assistant...",
        "model": "gemini-2.5-flash"
      }
    ],
    "edges": []
  }
}

---

Flow Types

Sequential

Agents execute one after another, each receiving the previous agent's output.

flowchart LR
    A["Agent A<br/>Research"] --> B["Agent B<br/>Analysis"] --> C["Agent C<br/>Report"]

Best for: Linear pipelines where each step depends on the previous step's output.

Example: Research → Analyze → Generate Report

Parallel

Multiple agents execute simultaneously, and results are merged.

flowchart TB
    Input["Input"]
    A["Agent A<br/>Market Analysis"]
    B["Agent B<br/>Financial Analysis"]
    C["Agent C<br/>Risk Assessment"]
    Merge["Merge Results"]

    Input --> A & B & C
    A & B & C --> Merge

Best for: Independent analyses that can run concurrently.

Example: Multi-perspective market research

Supervisor

A supervisor agent delegates work to specialist agents and synthesizes their outputs.

flowchart TB
    Supervisor["🧠 Supervisor<br/>Coordinates & synthesizes"]
    S1["Specialist 1"]
    S2["Specialist 2"]
    S3["Specialist 3"]
    Output["Consolidated Output"]

    Supervisor --> S1 & S2 & S3
    S1 & S2 & S3 --> Supervisor
    Supervisor --> Output

Best for: Complex multi-domain problems requiring cross-domain synthesis.

Example: ABA therapy goal generation, comprehensive market assessment

Dynamic

Agents decide at runtime which agent to call next based on intermediate results.

flowchart TB
    Start["Start"]
    Triage["Triage Agent"]
    Tech["Technical Support"]
    Billing["Billing Support"]  
    Escalate["Escalation Agent"]
    Resolve["Resolution"]

    Start --> Triage
    Triage -->|"Technical issue"| Tech
    Triage -->|"Billing issue"| Billing
    Tech -->|"Needs escalation"| Escalate
    Billing -->|"Needs escalation"| Escalate
    Tech -->|"Resolved"| Resolve
    Billing -->|"Resolved"| Resolve
    Escalate --> Resolve

Best for: Adaptive workflows with conditional branching.

Example: Customer support triage, diagnostic troubleshooting

---

Executing Workflows

Via UI

1. Open a workflow from the workflow list
2. Provide the required input (text or structured data)
3. Click Execute
4. Monitor real-time progress as each agent runs
5. View the consolidated output

Via API

POST /api/v1/workflow/execute
Content-Type: application/json

{
  "workflowId": "workflow-id-here",
  "input": {
    "clientProfile": {
      "name": "John Doe",
      "age": 7,
      "diagnosis": "ASD Level 2"
    }
  }
}

The response streams execution events via SSE:

event: agent_start
data: {"agentId": "aba-specialist", "status": "RUNNING"}

event: agent_output
data: {"agentId": "aba-specialist", "output": "..."}

event: workflow_complete
data: {"status": "COMPLETED", "output": "..."}

---

Workflow Templates

Synaptiq includes pre-built templates for common patterns:

Template	Flow Type	Description
Research & Report	Sequential	Research → Analyze → Generate Report
Multi-Perspective Analysis	Parallel	Multiple analysts → Consolidated view
Expert Panel	Supervisor	Coordinator manages domain experts
Support Triage	Dynamic	Auto-routes to appropriate specialist
Content Pipeline	Sequential	Draft → Edit → Review → Publish

Access templates: GET /api/v1/workflow/templates

---

Run History

Every workflow execution is recorded with:

• Run ID — unique identifier
• Status — RUNNING, COMPLETED, FAILED
• Timestamps — start, end, duration
• Agent outputs — individual results from each agent
• Consolidated output — final merged result

View runs: GET /api/v1/workflow/{workflowId}/runs

View run detail: GET /api/v1/workflow/runs/{runId}