Opidocs
Guides

Build a custom agent

Create a specialized AI employee that handles a specific job in your workspace.

What you'll build

By the end of this guide, you'll have a fully configured AI agent (called an "AI Employee" in Opisense) that's tailored to a specific role — like a customer support rep, content writer, or data analyst. You'll give it a name, personality, tools, and knowledge, then test it with a real conversation.

Prerequisites

  • You've completed the onboarding wizard and have an active workspace
  • You have an idea of what job you want your agent to handle (don't worry, we'll help you shape it)

Steps

Go to AI Employees

Click AI Employees in the sidebar. This is your agent hub — it shows all the AI employees you've created along with their status and recent activity. If you haven't created one yet, you'll see a prompt to get started.

Choose your creation path

Click Create and decide how you want to build your agent. You have two options:

  • Start from scratch — Full control over every setting from the beginning
  • Use a template — Pick a pre-built role (like "Support Agent" or "Research Analyst") and customize from there

Templates are a great starting point, especially if you're building your first agent. You can always change every setting after creation — nothing is locked in.

Name and describe your agent

Give your agent a clear name that reflects its role — something like "Support Bot" or "Content Reviewer." Then write a short description of what this agent does. This description helps your team understand the agent's purpose at a glance.

Configure the personality

This is where you shape how your agent communicates. You'll set:

  • Tone — Formal, casual, friendly, technical, or somewhere in between
  • Role context — Background information about what the agent should know about its job
  • Instructions — Specific guidelines for how the agent should behave, what to prioritize, and what to avoid

Think of this as writing a job description. The more specific you are, the more consistent and useful your agent's responses will be.

Add tools and skills

Tools extend what your agent can do beyond just generating text. Depending on the role, you might enable:

  • Web search — For agents that need to look up current information
  • Document processing — For agents that work with uploaded files
  • Data analysis — For agents that need to crunch numbers or interpret data
  • Custom skills — Specialized capabilities built for your specific workflows

Pick only what your agent actually needs. More tools doesn't always mean better — a focused agent with the right tools outperforms a generalist every time.

Connect knowledge sources

Link your agent to relevant documents from your knowledge base. This gives the agent context about your business, processes, or domain. For example, a support agent might need access to your product documentation, while a content writer might need your brand guidelines.

Agents can only access knowledge you explicitly connect. If your agent gives answers that seem uninformed, check that the right documents are linked.

Test with a conversation

Once everything is configured, it's time to take your agent for a spin. Click Test to open a chat with your new agent. Try a few different prompts that match the agent's role:

  • Ask it a question it should be able to answer from its knowledge
  • Give it a task that uses one of its tools
  • Test edge cases to see how it handles things outside its scope

If the responses aren't quite right, go back and tweak the personality, instructions, or knowledge connections. Building a great agent is an iterative process.

What's next?

Your custom agent is ready to work. Here's where to go from here:

  • Share with your team — Other workspace members can interact with your agent from the AI Employees page
  • Dive deeper into agent concepts — Read Agents and Skills to understand how agents, skills, and knowledge work together
  • Explore the full feature set — Check out the AI Agents reference for advanced configuration like analytics, execution history, and fine-tuning

On this page