Create Your First AI Flow

AI Flows are how organizations transform AI capabilities into repeatable business processes.

Unlike a simple prompt, an AI Flow can:

• Accept structured inputs
• Execute AI reasoning
• Use Skills and MCP tools
• Trigger sub-flows
• Produce structured outputs
• Integrate into larger workflows

In this guide, we’ll create an Incident Analysis Flow that analyzes production incidents and generates structured summaries.

---

Video Tutorial

Follow the complete walkthrough to create your first AI Flow in Karyam.

---

How AI Flows Work

Input Parameters
        ↓
System Instructions
        ↓
User Prompt
        ↓
AI Reasoning
        ↓
Skills & MCP Tools
        ↓
Output Formatting
        ↓
Structured Result

---

Step 1 — Open the Flow Builder

Navigate to:

Workspace → AI Flows → Create New Flow

You will see the following sections:

• General Information
• Model & Tuning
• Skills & Tools
• Categories
• Inputs
• System Instructions
• User Prompt
• Output Format

---

Step 2 — Configure General Information

Name

Choose a business-oriented name that clearly describes the purpose of the flow.

Examples:

• Incident Analyzer
• Vendor Risk Assessment
• Meeting Summarizer
• Contract Reviewer
• Customer Feedback Classifier

Avoid:

• Flow 1
• Test Flow
• AI Workflow

---

Description

Explain what the flow does and where it will be used.

Example:

Analyzes production incidents and generates structured reports
including root cause, impact assessment, and recommended actions.

---

Icon

Choose an icon that visually represents the flow.

This makes large workspaces easier to navigate and organize.

---

Step 3 — Configure Model & Tuning

Model

Select the model that best matches the complexity of your workflow.

Use Case	Recommended Model
Classification	GPT-5 Mini
Summarization	Claude Sonnet
Deep Analysis	GPT-5
Cost-sensitive Tasks	Gemini Flash
Private Infrastructure	Ollama

---

Maximum Steps

Controls how many reasoning iterations the flow can perform.

Flow Type	Recommended Steps
Classification	5
Summarization	8
Analysis	10
Multi-stage Reasoning	15+

Higher values improve reasoning quality but increase latency and cost.

---

Temperature

Controls creativity.

Value	Behavior
0.0	Deterministic
0.2	Focused
0.5	Balanced
0.8	Creative
1.0	Highly creative

Recommended:

• Data extraction → 0.0
• Analysis → 0.2
• Content generation → 0.8

---

Enable Caching

Enable caching to improve response times and reduce costs for repeated requests.

Recommended for:

• Summarization flows
• Classification flows
• Frequently executed workflows

---

Step 4 — Attach Skills & Tools

Flows become significantly more powerful when connected to external systems.

Skills

Attach up to 5 Skills.

Examples:

• Send Email
• Search CRM
• Query Database
• Create Ticket

---

MCP Tools

Attach up to 5 MCP Tools.

Examples:

• GitHub
• Slack
• PostgreSQL
• Jira
• Salesforce

---

Sub-Flows

Attach up to 5 sub-flows.

This allows complex workflows to be broken into smaller reusable components.

Example:

Incident Analysis
├── Root Cause Analysis
├── Impact Assessment
└── Executive Summary

---

Step 5 — Organize Using Categories

Categories help organize large collections of flows.

Examples:

• Operations
• Finance
• HR
• Customer Support
• Engineering

You can assign up to 5 categories to each flow.

---

Step 6 — Define Inputs

Inputs allow flows to receive dynamic values at runtime.

Click Add Input Field to create an input.

Examples:

Name	Type
incident_id	Text
customer_name	Text
severity	Select
region	Text

---

Using Inputs in Prompts

Inputs can be referenced using placeholders:

Analyze incident ${incident_id}
for customer ${customer_name}
with severity ${severity}.

At runtime these placeholders are automatically replaced with real values.

---

Step 7 — Configure System Instructions

System Instructions define:

• Role
• Behavior
• Constraints
• Decision boundaries
• Tone

These instructions run before every execution.

Example:

You are a Senior Site Reliability Engineer.

Responsibilities:
- Analyze incidents.
- Identify probable root causes.
- Recommend mitigation actions.

Rules:
- Never speculate without evidence.
- Prefer data-driven conclusions.
- Return concise technical explanations.

---

Step 8 — Configure the User Prompt

The User Prompt defines the task the flow performs.

Unlike System Instructions, this prompt changes between flows.

Example:

Analyze the following incident:

Incident ID: ${incident_id}
Severity: ${severity}
Region: ${region}

Provide:

1. Root Cause
2. Impact Assessment
3. Recommended Actions

---

Step 9 — Define Output Format

Output Formats allow flows to return structured results.

Examples:

JSON

{
  "root_cause": "",
  "impact": "",
  "actions": []
}

---

Markdown

## Root Cause

## Impact

## Recommendations

---

Table

Field	Value
Root Cause
Impact
Actions

If left empty, the AI will return its default output format.

---

Example Configuration

Incident Analysis Flow

Setting	Value
Model	GPT-5
Steps	10
Temperature	0.2
Skills	Jira Search
MCP Tools	PostgreSQL
Inputs	incident_id, severity
Output	JSON

---

Best Practices

Keep Flows Focused

Prefer smaller reusable flows over large monolithic workflows.

---

Use Inputs Instead of Hardcoding

Dynamic inputs make flows reusable across multiple scenarios.

---

Structure Outputs

JSON outputs are easier to consume in downstream workflows and automations.

---

Use Sub-Flows

Break large business processes into smaller reusable building blocks.

---

Enable Observability

Monitor:

• Flow Logs
• Execution Traces
• Token Usage
• Cost Analytics

Production AI requires visibility.

---

Your First Flow Is Ready

Congratulations 🎉

You have created your first AI Flow in Karyam.

Next, connect your flow to Agents, Skills, and business systems to automate real work.

---