Building Your First AI Agent with AzelaAIOS

Building Your First AI Agent

This guide walks you through the complete process of building a production-ready AI agent in AzelaAIOS — from deciding what the agent should do to deploying it and monitoring its runs.

Step 1: Define the Goal

Before opening the Agent Builder, write a one-sentence goal for your agent. Example:

"Research our top 10 competitors and produce a weekly positioning report for the product team."

The clearer your goal, the better your agent will perform.

Step 2: Open the Agent Builder

From the AzelaAIOS dashboard, click New Agent and select Build from scratch. You can also start from one of the 40+ marketplace templates if your use case is common.

Step 3: Set Role and Basic Info

Give your agent a descriptive name and select its role type (Research, Operations, Customer Support, etc.). This helps the system pre-configure defaults for model selection and tool access.

Step 4: Select the AI Model

Choose from:

• GPT-4o — Best for complex reasoning and long-form outputs
• Claude 3.5 Sonnet — Excellent for analysis and structured outputs
• GPT-4o Mini — Fast and cost-efficient for simpler tasks
• Auto — AzelaAIOS selects the best model per task automatically

For a research agent, Claude 3.5 Sonnet or GPT-4o are strong choices.

Step 5: Attach Tools and Connectors

Select the connectors your agent needs:

• Google Search for web research
• Google Drive for reading internal documents
• Notion for writing reports
• Slack for sending notifications

Connectors are pre-built — just authorise the connection and the agent can use them immediately.

Step 6: Add Knowledge Bases

Upload PDFs, strategy documents, previous reports or any reference material. The agent will retrieve relevant content from these during execution.

Step 7: Configure Memory

Enable cross-run memory so the agent can reference what it found last week. Choose between:

• Structured facts — key-value pairs the agent updates over time
• Semantic memory — vector-indexed text the agent can query contextually

Step 8: Set Instructions and Guardrails

Write the system prompt that governs how the agent thinks and responds. Set guardrails to prevent:

• Off-topic outputs
• Hallucinated citations
• Actions the agent should never take autonomously

Step 9: Test in the Testing Lab

Run the agent in the Testing Lab with a sample prompt. Review the reasoning trace, tool calls and final output. Adjust instructions if needed.

Step 10: Deploy

Once satisfied, deploy the agent as:

• A scheduled run (e.g. every Monday at 8am)
• An API endpoint your team can call programmatically
• An embedded widget in your internal tools

All runs are logged, and any flagged actions require approval before execution.

What Happens After Deployment

Every agent run generates a full execution trace — which tools were called, what data was retrieved, what was written and which approval gates were triggered. You can view this in the Monitoring dashboard and set up alerts for failures or anomalies.

Build Your First Agent →