Try Agent Mesh
In this tutorial, you build a multi-agent system for a fictional coffee shop, Solly's Coffee. Your agents check inventory levels, pull product data from a supplier's REST API, generate purchase orders when stock runs low, and respond automatically when events arrive on an event broker.
Along the way, you'll see how Agent Mesh handles the hard parts of agent coordination: routing requests to the right specialist, connecting agents to external data via connectors, enforcing consistent business processes, validating agent behavior before deployment, and exposing agents to external systems through entrypoints. The tutorial covers both dynamic orchestration, where a large language model (LLM) decides which agents to call, and deterministic workflows, where the execution path is prescribed in code.
This tutorial follows the agent development lifecycle (ADLC), a framework for building production-ready agents:
-
Build — Create and configure agents.
-
Test — Validate agent behavior with evaluations.
-
Deploy — Push configurations to production environments.
-
Observe — Monitor agent performance and interactions.
-
Improve — Refine agents based on production feedback.
Prerequisites
This tutorial uses both the Agent Mesh Desktop application and the Agent Mesh CLI. Desktop runs locally and connects to an LLM provider for AI capabilities; the Agent Mesh CLI deploys your configurations to it.
Before you begin, make sure you have the following:
-
A supported operating system — Windows 10+, macOS 11+, or Linux
-
LLM API credentials — from Anthropic, OpenAI, Google, Azure, AWS, or a local Ollama installation
-
Docker or Podman — required for Step 7 only, to run a local event broker instance
-
60-90 minutes — to complete the tutorial
Tutorial Steps
Complete the steps in order. Each step builds on agents and configurations created in earlier steps.
| Step | Description |
|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|