Building an AI Agent from Scratch: A Complete Guide from Soul Files to Cloud Deployment

Many people want their own AI agent, but the moment they open any tutorial, the technical complexity sends them running. This article breaks down a beginner-friendly end-to-end setup — from the often-overlooked Soul file to cloud deployment and skill automation — helping you get a genuinely functional AI agent up and running with the lowest possible technical barrier.

The Key Most People Miss: The AI Soul File

When beginners complain about poor AI agent output quality, their first instinct is to tweak their prompts. But the real problem is often more subtle — it lives in what developers call the "Soul" configuration file.

Every time you open a new conversation, whether with a basic chatbot or an AI agent, the system loads a hidden document in the background. This document defines how the AI should behave, how it should respond, and a set of behavioral rules. That's the Soul file. The catch is that even if you know it exists, writing a good one is genuinely difficult — how to structure it, what to put in each section — all of that takes accumulated experience.

A free tool called Ordain is worth recommending here — it automatically generates high-quality Soul files for you. When you open the tool, you'll see two core options: Soul (the Soul file) and Skill. The Soul file contains several configuration blocks:

• Agent selection: Beginners can go straight with Hermes
• Agent name: Customize as you like
• Task scope: Multi-select options like sending messages and emails, research and summarization, etc.
• Personality: Direct, warm, playful, or professional
• Autonomy level: This is the most critical setting. You can have it ask for your approval before doing anything, or let it operate fully autonomously. The recommended setting is "handle small things independently, check in for big decisions" — so you're not bothered by trivial tasks, but retain control over important choices

Once you've filled everything in, click "Generate soul.md" and the tool produces a detailed Soul file in seconds — covering modules like "who you are," "what you're responsible for," "what you must never do," and "memory." A document that would take over an hour to write by hand is generated instantly. You can also add extra content in the custom field and click "refine" to iterate at any time.

Cloud Deployment: Bypassing the Most Complex Technical Hurdles

This is the step where most tutorials start losing beginners. Running an AI agent typically means downloading multiple files, checking for security issues, and running a series of terminal commands — for someone who's never done it before, that's easily a few hours of frustration.

On top of that, running an AI agent on your local laptop is a bad idea: the moment you close the lid and the computer sleeps, the agent goes offline too. Most people opt for cloud server deployment — but that involves a whole new set of technical knowledge, from unfamiliar operating systems to SSH connections.

A beginner-friendly solution is Hostinger, which offers one-click deployment for agents like Hermes. Here's the process:

1. At the bottom of Ordain, click "Install Hermes on Hostinger"
2. Choose a plan — KVM2 is recommended, with 8GB of RAM that's enough to push the agent to its performance ceiling without waste or bottlenecks
3. During deployment, you'll be shown a username and an auto-generated password — save the password, you'll need it for later configuration
4. Click Deploy, and Hostinger will automatically build the server and install Hermes

Once deployment is complete, you only need to paste three commands in the terminal: open the Hermes folder, start Hermes itself, and run Hermes Setup to launch the configuration wizard. Almost no low-level technical work is involved.

Connecting Your AI Model and Messaging Channel

The configuration wizard will walk you through a few key settings. First is the AI model provider — OpenRouter is recommended because it connects to virtually every major LLM on the market, and switching between models takes just one command.

Create a free OpenRouter account and generate an API key. One important note: always set a spending limit (e.g., $25/month) to avoid waking up one day to find hundreds of dollars charged to your account. After adding funds, you can choose which model to run — OpenRouter offers a range of models with good price-to-performance ratios, and some free models for users on a tight budget.

Next, configure your messaging channel. The wizard supports multiple platforms; Telegram is the recommended choice for its simple setup. You'll need two things:

• Bot Token: Search for BotFather on Telegram, send /newbot, give it a name and a username ending in "bot," and you'll receive your token
• User ID: Search for userinfobot, send start, and you'll get your user ID

After entering both into Hermes, you'll also need to start a Gateway — it maintains the persistent connection between Hermes and Telegram and listens for your messages in real time. This step is critical; without it, the entire agent system won't function properly.

Putting Your AI Agent to Work: Task Execution and Memory

At this point the agent can respond to messages, but it's still in a generic state because the Soul file is still sitting on your local machine. Here's a key detail: you can't just upload the file — the AI will only be able to guess at what it means. The correct approach is to go back to Ordain, click "copy prompt" below the generated file, copy the accompanying instructions, then paste those instructions into Telegram along with the Soul file itself.

Once configured, give the agent a real, multi-part task to test it. For example: research online the "women over 30 who want to lose fat through strength training" audience, compile the findings and write five UGC-style scripts, then search for the best AI video generation tools for UGC ads and their pricing.

A few minutes later, the agent will automatically break down the task — do the research first, write the scripts, then complete the tool comparison — all triggered by a single Telegram message. This is the fundamental difference between an AI agent and a regular chatbot: it can autonomously execute multi-step tasks and deliver a final result without requiring you to intervene at every step.

Crucially, thanks to the memory module in the Soul file, the agent also has persistent memory. Once you tell it about your target audience, it will remember that context in future sessions — even after starting a new conversation.

Using Skills to Automate Tasks

What gives an AI agent real long-term value is the Skill mechanism. Imagine you want to receive the latest industry insights about "AI UGC ads" every morning at 9 AM, with strict rules: only search content published in the past 28 days, only keep actionable practical advice, prioritize highly relevant sources, and remove duplicate information.

In Ordain, select the platform (Hermes), name the skill, paste in the rule descriptions, and click "Generate skill." The tool produces two configuration files. Using the same "copy prompt + attachment" approach, send them to Telegram, and Hermes will automatically install the skill.

The final step is scheduling. If you have to manually trigger the skill every time, that defeats the purpose of an agent working autonomously around the clock. In Hermes, you simply tell it in plain language to "run this skill every morning at 9 AM," and the task enters the scheduler to execute automatically at that time every day — a true hands-free automated workflow.

Summary: Key Takeaways for Building an AI Agent

The value of this approach isn't any single tool — it's that it takes AI agent setup, which previously required developer-level skills, and reduces it to a handful of steps that anyone can follow:

• The Soul file determines output quality and memory capability — it's worth investing time here first
• Cloud deployment eliminates all the pain points of local environments, keeping your agent online 24/7
• When connecting your model, always set an API spending limit to control costs
• Skills + scheduled automation are what make an agent truly "work on its own"

One final note: the specific tools mentioned in this article (Ordain, Hostinger, Hermes) are recommendations for one particular setup. When deploying for real, evaluate based on your own needs — and pay attention to API spending limits, data security, and service provider reliability.