Structured Outputs

In Lyzr, Structured Outputs allow an agent’s response to be returned in a machine-readable format such as JSON, YAML, or even tabular data. These are crucial when your agent needs to interact with other systems, APIs, databases, or dashboards.

Instead of returning free-form text — which is flexible but difficult to extract meaning from — structured outputs make the agent’s responses actionable, testable, and easier to plug into production-grade workflows.

---

✨ Why Structured Outputs Matter

✅ Machine Interoperability

Structured outputs can be directly consumed by APIs, connected to automation tools like Zapier, integrated into frontend dashboards, or stored in structured databases like PostgreSQL or MongoDB.

Example: Imagine a marketing agent returning campaign stats:

{
  "campaign_name": "July Blast",
  "clicks": 4231,
  "conversions": 372,
  "budget_remaining": 129.5
}

This response can be fed into an analytics dashboard or automation rule.

✅ Predictable Parsing

Instead of using regular expressions or NLP techniques to extract meaning from agent output, you always know what structure to expect. This drastically reduces the complexity of downstream logic.

✅ Enhanced UI Rendering

You can automatically render JSON objects as tables, charts, forms, or timelines in the UI — useful for dashboards, forms, reports, and more.

✅ Validation & Error Detection

You can define strict schemas for agent responses — including required fields, data types, and constraints. This enables robust validation and early detection of hallucinations or formatting mistakes.

---

🧠 Use Cases with Real Examples

🎯 Lead Scoring Agent

{
  "lead_name": "Jane Doe",
  "lead_source": "LinkedIn",
  "score": 91,
  "confidence": "High"
}

🛠️ Tech Support Ticket Generator

{
  "issue": "Password reset not working",
  "urgency": "High",
  "customer_id": "CUST-29387",
  "assigned_to": "[e-mail]"
}

📬 Email Generator for Sales Outreach

{
  "subject": "Re: Interest in Data Platform",
  "body": "Hi Mark,\nJust following up on our previous conversation...",
  "call_to_action": "Schedule a demo"
}

📊 Data Summary Agent

{
  "summary": "Revenue increased by 15% QoQ",
  "top_regions": ["India", "US", "UK"],
  "growth_drivers": ["Product X", "Campaign Y"]
}

---

🏗️ How to Use Structured Outputs in Lyzr Studio

1. Agent Configuration

When creating or editing an agent in Lyzr Studio:

• Scroll to the Expected Output Format section.
• Toggle Structured Output to ON.

2. Define Output Structure

Specify a sample output format or a JSON schema.

Example Output Format:

{
  "name": "string",
  "email": "string",
  "score": "integer",
  "comments": "string"
}

3. Prompt Engineering

Embed structure instructions directly in your prompt:

Return the result in the following JSON format. Do not include any explanation or markdown:
{
  "issue": "<string>",
  "priority": "<Low|Medium|High>",
  "assignee": "<email>"
}

4. Enable Validation (Optional)

You can enable schema validation within Lyzr to:

• Reject malformed responses
• Retry generation
• Trigger fallback behavior or alerts

---

✅ Best Practices

Tip	Description
💬 Be explicit	Tell the agent exactly what structure to follow, with a JSON snippet.
🔄 Add retries	On schema mismatch, set up retries with more specific prompts.
✂️ Avoid overkill	Only include necessary fields — don’t make structures too verbose.
🧪 Test edge cases	Try inputs like nulls, large numbers, or ambiguous prompts.
⚙️ Define schemas	Use JSON Schema for validation when needed.

---

🚫 When Not to Use Structured Outputs

• When you’re building conversational, storytelling, or creative agents.
• When the output is unpredictable, such as philosophical insights or user interviews.
• When human readers are the only consumers of the agent’s response.

Use free-form text in those cases.

---

📌 Summary

Attribute	Value
Format	JSON, YAML, key-value pairs, tables
Ideal for	Automation, dashboards, pipelines, data pipelines
Setup	Agent creation > Enable Structured Output
Prompting Style	Strict instruction + Output structure
Integration Use	High — works with webhooks, APIs, databases, frontend UIs

---

🔚 Final Note

By using Structured Outputs in your Lyzr agents, you’re enabling automation, reducing complexity, and building reliable integrations. Treat your agents not just as conversational entities — but as programmable, structured interfaces into your systems.