Build Your First AI Agent in 10 Minutes

TL;DR: You can build a working AI agent in about 10 minutes on Alysium — no coding, no technical background, no developer required. You'll need your existing content (a document, a list of FAQs, anything written down), a free account, and 10 focused minutes.

Every week, someone tells me they "need to learn to code" before they can build an AI product. They don't. They never did.

Building a custom AI agent today means uploading your existing knowledge, writing a few paragraphs of instructions, and clicking publish. The entire workflow — from blank screen to live agent — takes about 10 minutes. No programming. No API keys. No technical team. If you can write an email, you can do this.

This is that walkthrough.

Step 1: Create Your Free Account

Head to Alysium and create an account with your email and password. The signup process takes under a minute.

Once you're in, you'll see the Chat Hub — your central dashboard where all your agents live. It's empty right now. By the end of this guide, it won't be.

Expected outcome: You're logged in and looking at the Chat Hub. No agents yet.

One thing to know before you start: Alysium's free tier doesn't require a credit card. You can build, test, and iterate your full agent without entering payment information. The paid tier only becomes relevant when you want to scale usage beyond the free allowance — which most first-time builders don't hit during the build-and-test phase. Sign up, verify your email, and you're in.

Step 2: Create a New Agent

Click the button to create a new agent. You'll be taken to the agent builder — the control center for everything your agent knows and how it behaves.

The first thing you'll do is give your agent a name. Make it descriptive. If this is an FAQ agent for your photography business, something like "Studio FAQ" is more useful than "My Agent." Visitors will see this name, so it's worth a moment of thought.

You can also write a short description here — this appears on agent cards and helps visitors know what they're getting before they start chatting.

Expected outcome: Your agent has a name and description. The builder is ready for configuration.

The name and description fields matter more than they seem. The name appears in the widget header and in any marketplace listing — it should be something a user would recognize as relevant to their question. The description is mostly internal at this stage, but writing it carefully forces clarity: if you can't describe what the agent does in two sentences, the instructions will probably be fuzzy too. Use the description as a forcing function for scope.

Step 3: Write Your Instructions

This is the most important step. The instructions field — which holds up to 8,000 characters — is where you define your agent's entire personality, scope, and behavior. Think of it as the briefing you'd give a knowledgeable assistant before their first day.

Here's a simple structure that works:

• Who the agent is: "You are an FAQ assistant for [Business Name]. You help customers find information about our products, pricing, and policies."
• Tone: "Respond in a friendly, professional tone. Use short paragraphs. Avoid jargon."
• Boundaries: "Only answer questions based on the information in your knowledge base. If you don't know, say so honestly and suggest the visitor contact us directly."
• What to avoid: "Don't make up pricing, hours, or any other factual claims you can't verify from the uploaded documents."

You don't need all 8,000 characters for a basic agent. Two or three focused paragraphs is a solid start — you can expand later.

Expected outcome: Your agent has a working set of behavioral instructions.

Step 4: Upload Your Content

The knowledge base is what separates your agent from a generic chatbot. This is where you give it the information it will draw from when answering questions.

Alysium supports 11 file types: PDF, Word documents (.doc, .docx), Excel spreadsheets (.xls, .xlsx), PowerPoint presentations (.ppt, .pptx), plain text (.txt), Markdown (.md), CSV, and HTML. You can also paste text directly — useful for SOPs, FAQs, or anything living in a Google Doc.

What should you upload? Start with the most common questions you get. If you're a coach, that might be your intake FAQ, a methodology overview, or a client handbook. If you're a small business owner, it might be your price list, hours, and service descriptions.

Documents process in the background with a live status indicator. By the time you've finished the next steps, they'll be indexed and ready.

A few tips on what works well for a first upload: FAQ documents and service descriptions tend to produce strong out-of-the-box results because they're already written in question-and-answer format — which maps naturally to how users interact with agents. Long, unstructured documents like transcripts or meeting notes tend to produce weaker results on a first pass. If you have both, start with the structured material.

Expected outcome: At least one document uploaded and processing. Status shows "indexed" or similar.

Step 5: Set Up Conversation Starters

Conversation starters are the suggested prompts that appear on your agent's welcome screen — they guide visitors toward the most useful questions. You can add up to 5.

Good starters are specific, not generic. "Tell me about your services" is weak. "What's included in the $299 headshot package?" gives visitors a clear picture of what the agent can actually help with — and pulls them into a real conversation faster.

You can add an emoji to each starter to make the welcome screen more inviting. Not required, but they do make the experience feel warmer.

Expected outcome: 3–5 conversation starters written. Your agent's welcome screen is starting to look like something real.

The best conversation starters are the questions you're already tired of answering. Look back at the last 20 messages you've received from clients, students, or customers — the questions that show up repeatedly are exactly the right starting points. Phrase them the way users would actually type them, not the way you'd formally categorize them. "How does the program work?" performs better than "Program Overview" because users think in questions, not categories.

Step 6: Customize the Welcome Message

When a visitor opens your agent, the first thing they see is the welcome message. This is your introduction — a brief greeting that sets the tone and tells them what the agent can help with.

Keep it short. Something like: "Hi! I'm [Agent Name], your guide to [topic]. Ask me anything about [topic A], [topic B], or [topic C]."

This isn't a legal disclosure or a terms page. It's the first impression. Make it warm.

Expected outcome: A working welcome message that matches your agent's purpose.

Think of the welcome message as the difference between a door with a sign that says "help available" versus one that says "ask me anything about [specific topic]." The second one converts more visitors to first questions because it's specific about what kind of help is available. A good pattern: one sentence on what the agent knows, one on what kinds of questions it's best suited for, and one optional line setting expectations ("I'll tell you if I don't know").

Step 7: Choose a Widget Theme (Optional but Worth It)

If you plan to embed this agent on a website, Alysium's widget designer gives you full control over how it looks. There are 36 built-in themes — everything from Light and Dark to Synthwave, Retro, and Cyberpunk. You can also override the accent color with any hex value or add custom CSS.

The live preview panel shows every change in real time, so you can see exactly what your visitors will see before you publish.

If you're just testing right now and don't have a website in mind, skip this step. You can always come back to it.

Expected outcome: A widget appearance you're happy with — or a mental note to come back to this step later.

Theme selection is worth ten minutes even if you plan to customize later. The 36 built-in options vary by visual weight and color temperature — lighter themes work better on content-heavy pages where you want the widget to feel integrated, darker themes work better as standalone tools where you want the agent to feel premium. The custom hex accent color is the most efficient single change: matching your brand's primary color creates a sense of intention that distinguishes a custom-built agent from a generic chatbot.

Step 8: Publish and Share

Hit the publish button. Your agent goes live immediately.

Alysium generates a shareable link you can send to anyone — no Alysium account required for the recipient. You'll also get a script tag you can paste into any website to embed the floating chat widget. If you want to restrict embedding to specific domains, you can configure that too.

Before you share, do one quick test yourself: ask a question your content should be able to answer. Then ask a question it definitely can't answer — and verify your agent handles it gracefully ("I don't have that information — please reach out to us directly" is the right response, not a made-up answer).

Also try phrasing the same question a few different ways. Real users rarely phrase things the way you'd expect — that variation test will tell you a lot about how solid your agent is before you send it to anyone who matters.

Expected outcome: Your agent is live, published, and you've done at least one test conversation.

What Happens Next?

Congratulations — you built an AI agent. Here's what most people discover within the first week:

The first version is never the last version. After real visitors use it, you'll see which questions it handles well and which ones expose gaps in your knowledge base. Upload more content. Refine your instructions. Add conversation starters that match how people actually phrase things.

Alysium's analytics dashboard shows you conversations, unique users, and helpfulness ratings by agent — so you have real data to work with, not guesswork.

Pay particular attention to the conversation history view. Reading through real conversations — even the first five or ten — will surface questions you didn't anticipate, phrasing patterns that trip up the agent, and topics that users care about that you haven't fully covered yet. That feedback loop is worth more than any configuration change you could make blind.

If you want to go deeper: learn how to train your AI on your content so it sounds like you, or read the guide to writing better instructions. Both will meaningfully improve what you just built.

Ready to build? Start free on Alysium — no credit card required.

The biggest improvement most builders make in week two is their knowledge base, not their instructions. After seeing real conversations, it becomes obvious which questions the agent struggles with — usually because the uploaded content doesn't address them directly enough. The fix is almost always the same: find the gap, write a document that addresses it specifically, upload it, and test again. Three rounds of this loop typically transforms a "good enough" agent into one you'd confidently put your name on.