Andrey Koptelov Showed How to Deploy an AI Agent in 8 Hours: From Idea to Working Prototype with Zero Budget

Imagine you're tasked with implementing artificial intelligence, but with no budget, a strict requirement to keep all data in-house, and an urgent deadline. This very common scenario is what Andrey Koptelov, a business process expert, addresses by demonstrating how to create a fully functional AI agent prototype in just 8 working hours. The result: a local language model integrated with an orchestrator, capable of analyzing websites, disseminating information, and even answering questions from an internal knowledge base, all without sending data externally.

AI implementation often seems like a complex and expensive project, requiring months of preparation and massive budgets. This is a myth that prevents companies from even starting. In reality, most routine text-related tasks can be automated with an AI agent in a matter of days, using readily available tools, proving the concept's effectiveness, and then scaling. The key is to start, not to wait for ideal conditions.

The Reality of the Problem: AI Needed Yesterday, But No Resources

Many companies face a paradox: AI technologies are becoming mainstream, but budgets for their implementation are limited or nonexistent. The IT department is overburdened, while the need to automate text-processing routines grows. A typical scenario involves employees spending hours on monotonous tasks: analyzing large volumes of information, compiling reports, searching for data in disparate sources, and answering standard inquiries. Moreover, data is often confidential and cannot be sent to external cloud services.

An additional constraint is team competency. Most employees know how to use chatbots but lack programming skills and the ability to deploy complex systems. In such circumstances, traditional approaches to AI implementation are not viable. A path was needed that would allow for quick creation of a working prototype, proving its value, and only then, if necessary, scaling it.

The Path to an AI Agent: From Chat to Automation

For a long time, the primary way most employees interacted with AI was through chat mode with a large language model. This is convenient for getting quick answers or generating texts but does not solve the problem of automating repetitive business processes. Companies needed a tool that could not just generate text, but perform a sequence of actions, make data-driven decisions, and integrate with other systems.

This is why Andrey Koptelov arrived at the concept of an AI agent: not just a chatbot, but a system capable of autonomously performing tasks. The key was using open-source software, which allows the entire technology stack to be deployed locally, without cloud dependencies or costs. This solution is ideal for pilot projects where a hypothesis needs to be quickly tested and tangible results obtained.

How the AI Agent Was Designed in One Day

The AI agent was designed with strict constraints in mind: zero budget, local deployment, and no programming skills among end-users. The solution was based on two key components: a local large language model and a no-code orchestrator.

• Local LLM (Ollama). The choice fell on Ollama—an open-source platform that allows models like Qwen to run directly on a standard PC. This ensures data confidentiality and independence from external services.
• No-code Orchestrator (n8n). For connecting the LLM to business processes, n8n was chosen—a tool for automating workflows without code. It allows for visually creating chains of actions, integrating various services and AI models.

The core functionality of the AI agent included:

• Website Analysis. The AI agent could regularly scan specified web resources, extract relevant information, and summarize it.
• Information Dissemination. The collected data was automatically formatted and sent to interested parties.
• Internal Knowledge Base (RAG). For answering questions based on confidential internal documents, the Retrieval-Augmented Generation (RAG) approach was used. This allows the agent to query a local vector knowledge base and provide accurate answers without sending data externally.

A key principle was to create a prototype that "works" and demonstrates value, rather than an "ideal" solution. This allowed development time to be minimized.

Implementation: 7 Steps to a Working Prototype

The implementation process was broken down into 7 sequential steps, each with a clear success criterion. This enabled even untrained employees to quickly grasp the technology.

1. Deploying a Local LLM in Ollama. Installing Ollama and loading a model (e.g., Qwen 3.5 9b) onto a PC with 32 GB RAM.
2. Deploying n8n with Docker Desktop. Installing Docker and launching n8n for seamless integration and process management.
3. Integrating n8n and Ollama. Creating a workflow in n8n that calls the local LLM and uses it for data processing.
4. Writing Prompts. Developing effective prompts that guide the AI agent to perform specific tasks (analysis, summarization, classification).
5. Connecting RAG (optional). Creating a vector knowledge base and integrating it with the agent to answer questions based on internal documents.
6. Demonstrating the Prototype. Presenting the working solution to management for feedback and further development.

The entire process took 8 working hours. The main focus was on quickly obtaining a result and validating the concept, rather than creating a high-load or perfectly optimized solution. This helped avoid "analysis paralysis" and move directly to practical application.

Results: A Functional AI Agent in One Day

After 8 hours of work, a fully functional AI agent prototype was achieved, capable of solving real business problems.

The main result was the proof of concept: AI agents can be implemented quickly and without significant costs, even with strict limitations. This opens the way for further scaling and integration into more complex business processes.

How to Implement This in Your Company

Andrey Koptelov's case demonstrates that the entry barrier to AI has significantly lowered. If you have routine text-processing tasks and budget/confidentiality constraints, this approach can be an ideal starting point:

• Identify Routine Tasks. Start with processes where employees spend a lot of time analyzing, summarizing, or searching for text information.
• Deploy a Local LLM. Use Ollama to run models on your hardware, ensuring confidentiality and independence.
• Automate with a No-Code Orchestrator. n8n or similar tools will allow you to quickly connect the LLM to your business processes without programming skills.
• Use RAG for Internal Knowledge Bases. If you need to work with confidential documents, RAG will allow the agent to answer questions without sending data externally.
• Start with a Prototype. Don't strive for perfection immediately. The goal is to quickly get a working solution that proves the value of AI in your specific case.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager