Building and Deploying Production AI Agents

Building and Deploying AI Agents with n8n

A practical workshop on tool use, RAG, human approval, guardrails, and deployment with Karim Nabil, Senior AI Engineer at FPT Software.

An AI agent handling customer support needs access to accurate business information, tools for checking orders, and clear rules for when to involve a person. Once deployed, it also needs secure credentials, error handling, and visibility into its executions and costs.

In this four-hour workshop, Karim Nabil will guide you through building these capabilities in n8n. You’ll work through a customer support agent for an online store, starting with a conversational workflow and progressively adding business tools, document retrieval, human approval, and security checks.

The workshop follows the project through deployment, with demonstrations of both n8n Cloud and self-hosting using Docker. Along the way, Karim will explain the design decisions behind the workflow, how to investigate failures, and the operational responsibilities that come with each hosting approach.

                                                       Sunday, 13 December 2026 · 6:00–10:00 pm EET (Finland time, UTC+2)
                                                       30 seats · $39 early bird until 25 November · $49 after that.

What you’ll build

The workshop project is an AI support agent that can carry a conversation across multiple messages, retrieve order details through a tool, answer policy questions using company documents, and route a refund request for human approval.

You’ll see how these capabilities fit together in one workflow, including what happens when information is missing, a tool fails, or a request falls outside the agent’s permitted scope. The same project provides the basis for the deployment demonstrations.

What you’ll learn

By the end of the workshop, you’ll understand how to:

• Translate a business use case into an agent workflow with defined responsibilities and success criteria.
• Connect an LLM to n8n, configure instructions and conversation memory, and handle incoming messages.
• Give an agent access to APIs, business data, and reusable workflows through narrowly scoped tools.
• Use retrieval-augmented generation (RAG) to answer questions from company documents.
• Place human approval checks before sensitive actions.
• Apply input and output checks, protect credentials and customer information, and recognize prompt injection risks.
• Inspect executions, diagnose failed requests, and monitor model usage and costs.
• Deploy through n8n Cloud or a Docker-based setup and test the workflow in its live environment.

Workshop outline

1. Define the use case and architecture
Identify the support tasks the agent should handle, the information it needs, and the actions that require approval. Map the connections between the chat interface, model, tools, and business data.

2. Build the conversational workflow
Work with n8n triggers, nodes, credentials, and executions. Connect a language model, establish the agent’s instructions, add conversation memory, and implement a basic error response.

3. Connect tools and business systems
Extend the workflow with an order-lookup tool backed by an API or database. Organize reusable workflows and structure tool responses so the agent can interpret the results consistently.

4. Add document retrieval and human approval
Prepare company documents for retrieval and use relevant policy information to support answers. Build an approval step into the refund workflow so a person reviews the request before an action proceeds.

5. Introduce guardrails and security
Test out-of-scope requests and attempts to obtain private information. Explore input and output validation, restricted tool access, credential protection, and the risks of instructions embedded in user messages or retrieved documents.

6. Inspect, troubleshoot, and monitor
Trace a conversation through its execution history, identify a failed step, and examine how the workflow handles it. Discuss usage costs and basic execution limits when requests overlap.

7. Deploy with n8n Cloud and Docker
Walk through publishing the workflow on n8n Cloud and running a self-hosted instance with Docker. Cover credentials, environment configuration, persistent storage, HTTPS, live chat or webhook endpoints, and checks to perform after deployment.

8. Run the complete workflow and discuss next steps
Bring the components together in a complete support interaction, including an approval request and a failure scenario. Review the implementation and discuss questions about adapting it to other use cases.

The session combines live building, guided explanations, practical demonstrations, and Q&A, with a 15-minute break.

Who should attend

This workshop is for developers, AI and data practitioners, and automation professionals who want to build agents that connect to business systems. It is also relevant to technical founders and product builders evaluating n8n for customer support or internal automation.

You’ll get the most value if you’re comfortable with basic AI concepts and have some familiarity with APIs and JSON. The workshop introduces the n8n concepts used throughout the project.

Recommended preparation

To build along, have access to an n8n instance and an API key for an LLM provider supported by n8n. Familiarity with Docker and basic server concepts will help you follow the self-hosted deployment demonstration.

Prerequisites

Basic familiarity with AI tools, APIs, and JSON will help you follow the workshop; prior experience with n8n is helpful but not essential. To build along, you’ll need a laptop, access to n8n Cloud or a local n8n instance, and a valid API key with available credits for an LLM provider. Basic knowledge of Docker and servers will help with the self-hosted deployment section.

Your instructor

Karim Nabil is a Senior AI Engineer at FPT Software, where his work includes designing and deploying agentic AI workflows with n8n and automating business processes for enterprise clients.

His experience spans AI engineering, product management, and business development, bringing both implementation and business perspectives to the workshop. He also creates the n8n with Karim educational content and delivers practical training on AI automation and agentic systems.

Registration

The workshop is limited to 30 participants.

Book through 15 November 2026 for the $39 USD early-bird price. From 16 November, tickets are $49 USD, subject to availability.