Build Your First Agent
This guide walks you through creating an AI agent from scratch. By the end, you'll have a working agent that can answer questions based on your own content. No coding or AI experience needed.
What you'll learn
- What an AI agent is and how it works in Orxa
- How to create and name your first agent
- How to write good instructions that shape agent behavior
- How to choose the right AI model for your needs
- How to upload documents so your agent can answer questions
- How to test your agent before making it live
What is an agent?
An agent in Orxa is an AI assistant that you configure for a specific purpose. Think of it like hiring a virtual team member:
- You tell it what its job is (through instructions)
- You give it knowledge (documents, FAQs, website content)
- You connect it to tools (Gmail, Slack, WhatsApp, etc.)
- It handles conversations with your customers or team 24/7
For example, you could create a "Customer Support" agent that knows your product documentation and can answer customer questions via WhatsApp — automatically, around the clock.
Agents don't replace your team — they handle routine questions so your team can focus on complex issues. You can always take over a conversation from the AI at any time.
Create an agent
Here's how to create your first agent:
- Click Agents in the left sidebar menu
- Click the Create agent button in the top-right corner
- Enter a name — this is for your reference, so make it descriptive. Examples: "Customer Support", "Sales Assistant", "FAQ Bot"
- Optionally add a description explaining what this agent does
- Click Create
Your agent is now created in Draft mode. This means it exists but isn't active yet — no one can interact with it until you publish it later.
Don't worry about getting the name perfect. You can change it anytime from the agent's settings page.
Write instructions for your agent
Instructions tell your agent how to behave. This is the most important step — good instructions make the difference between a helpful agent and a confusing one.
On your agent's page, go to the Settings tab. Scroll down to the Agent Behavior section and write your instructions.
Here's a template to get started:
Your job is to:
- Answer questions about our products and services
- Be friendly, professional, and concise
- If you don't know the answer, say so honestly
- Never make up information
Important rules:
- Always respond in [your language]
- Keep responses under 200 words
- If someone asks about pricing, direct them to our sales team
Be specific. Instead of "be helpful", say "answer questions about our products using the knowledge base, and offer to connect the customer with a human if the question is about billing or refunds."
Choose an AI model
The AI model is the "brain" your agent uses to understand questions and generate responses. Orxa supports multiple models from different providers:
OpenAI GPT-4o
Best for general-purpose agents. Good balance of speed and accuracy. Recommended for most users.
Anthropic Claude 3.5
Best for complex reasoning, following detailed instructions, and longer conversations.
Google Gemini 1.5
Best for speed. Good for high-volume agents that need fast responses.
You can also adjust the Temperature setting:
- Lower (0.1–0.3) — more consistent, factual responses. Good for support agents.
- Medium (0.5–0.7) — balanced. Good for most use cases.
- Higher (0.8–1.0) — more creative and varied. Good for brainstorming or creative writing.
Not sure which model to pick? Start with GPT-4o at temperature 0.5. You can change it anytime without losing your agent's knowledge or settings.
Add knowledge to your agent
Knowledge is the information your agent uses to answer questions. Without knowledge, your agent can only give generic responses. With knowledge, it can answer specific questions about your business.
Go to the Knowledge tab on your agent's page and click Add Information. You have four options:
Upload Files
Upload PDF, Word (DOCX), or text files. Great for product manuals, policy documents, or guides. Orxa reads and indexes the content automatically.
Write Text
Type or paste content directly. Good for company info, product descriptions, or internal policies that aren't in a document.
Q&A Pairs
Define specific question-and-answer pairs. Perfect for FAQs — the agent will give the exact answer you wrote when someone asks a matching question.
Website
Enter a URL and Orxa will read the page content. You can scrape a single page or crawl an entire website. Good for importing existing help center content.
After adding knowledge, you'll see it change from Processing to Indexed. This means Orxa has read, split, and organized the content so your agent can search through it when answering questions.
Content you add is processed by AI providers (like OpenAI) to create searchable indexes. Avoid uploading documents with sensitive personal information like ID numbers, passwords, or medical records.
Test your agent in the playground
The Playground is where you try out your agent before making it live. Think of it as a private chat room between you and your agent.
- Go to the Playground tab on your agent's page
- Make sure "Use Agent Knowledge" is turned on (the toggle at the top)
- Type a question a real user might ask — for example: "What are your business hours?"
- Check the response. Is it accurate? Is the tone right?
- Try edge cases: what happens when someone asks something your agent doesn't know?
Try toggling "Use Agent Knowledge" on and off. With it off, the agent only uses its general AI knowledge. With it on, it searches your uploaded content first. This helps you understand how much your knowledge base improves responses.
If responses aren't good enough, go back and:
- Improve your instructions (Step 3) to be more specific
- Add more knowledge (Step 5) to cover gaps
- Try a different AI model (Step 4) if the reasoning quality isn't right
Publish your agent
When you're happy with how your agent responds in the playground, it's time to make it live:
- Go to the Settings tab
- Find the Status dropdown
- Change it from Draft to Published
- Click Save
Your agent is now live. If you've connected integrations (like WhatsApp or web chat), it will start handling conversations automatically. You can monitor everything from the Inbox.
You can switch an agent back to Draft at any time to pause it. This stops it from handling new conversations, but doesn't delete anything.
Expected outcome
Congratulations! You've built and published your first AI agent. It now has instructions, knowledge, and is ready to handle conversations. Next, learn how to connect it to communication channels like WhatsApp, Gmail, or Slack.