Creating Your First Agent

Step-by-step guide to build, test, and publish an agent

This tutorial helps you build a reusable support assistant agent with optional tool connectivity and a clean version baseline.

What You’ll Build

• A project-scoped agent with system instructions
• Structured input/output variables
• Optional connector-backed tools
• A saved initial version ready for integration

Prerequisites

• Access to Bud AI Foundry
• Project with permissions to manage agents
• At least one available model/deployment

Step 1: Create the Workflow

1. Open the project Agents tab.
2. Click + Create New.
3. Confirm the workflow starts for the current project.

Step 2: Select Agent Type

1. Choose a prompt/agent type that matches your use case.
2. Continue to configuration.

Step 3: Configure Prompt

1. Select model/deployment target.
2. Add a concise system prompt.
3. Add role-based messages for expected interaction style.

Example system prompt:

You are a project support assistant. Provide concise, actionable guidance and call tools only when needed.

Step 4: Add Schemas (Recommended)

1. Enable structured input.
2. Add variables such as:
   • ticket_id (string)
   • priority (string)
3. Enable structured output with fields like:
   • summary (string)
   • next_actions (array)

Step 5: Connect Tools (Optional)

1. Open Tools.
2. Search/select connector.
3. Complete connector authorization.
4. Confirm it appears in connected tools.

Step 6: Run and Validate

1. Click Run Agent.
2. Provide sample input.
3. Verify response shape, clarity, and tool behavior.

Step 7: Save and Publish

1. Save the agent.
2. Confirm status in project table.
3. Open Use this agent and copy integration snippets.

Step 8: Prepare for Iteration

1. Open the agent detail page.
2. Go to Versions.
3. Create a new version for future improvements.

Start with minimal prompt scope, then expand after observing real traffic.