Skip to main content

Agents

Create custom AI agents with specific roles, knowledge sources, and tools—all without writing code

Updated this week

Building AI Agents on Hatz

This article covers:

  • Creating your first AI agent

  • Configuring agents with models, sources, and tools

  • Writing effective system prompts

  • Sharing agents with your organization

  • Using agents in Chat and Workshop

  • Best practices for agent development

What are AI Agents?

AI Agents are reusable, conversational chatbots with a specific persona, knowledge base, and set of tools. Unlike regular chat conversations, agents remember their instructions and can access specific sources and integrations every time you interact with them.

Agents are perfect for:

  • Marketing teams creating brand-consistent content

  • Sales teams managing prospect research and outreach

  • Operations teams automating administrative workflows

  • Finance teams handling data analysis and reporting

  • Any role where repetitive tasks can be automated

Your team members can focus on the work they were hired to do, while agents handle the routine admin work.

How to Create an Agent

Creating an agent is straightforward and intuitive:

  1. Navigate to the Workshop and click New

  2. Select Agent from the options

This takes you to the agent setup flow, where you'll configure your agent step-by-step.

Configuring Your Agent

Basic Information

  • Agent Name Give your agent a descriptive name that reflects its purpose. For example: "Brand Voice Marketing Agent" or "Sales Prospecting Assistant."

    • You can also give your agent a creative, personalized name if you prefer!

  • Description Add a brief description to help you keep track of what each agent does. This is especially useful if you're managing multiple agents.

    • Example: "Follows company brand guidelines and creates marketing content."

  • Category Select a category to make your agent easy to find later. Categories include Marketing, Sales, Operations, Finance, and more.

AI Model Selection

Choose the AI model your agent will run on. This works similarly to selecting a model in the chat interface.

Important: Make sure the model you select supports tool calling if your agent needs to use integrations or built-in tools.

Popular choices include:

  • Claude 4.5 Sonnet

  • Gemini 2.5 Flash

  • GPT-5

Model selection is largely based on personal preference, but consider the specific capabilities your agent will need.

Sources

Upload documents and files that your agent should reference. This could include:

  • Brand guidelines

  • Company policies

  • Product documentation

  • Style guides

  • Color palettes

  • Standard operating procedures

Sources can be uploaded directly or accessed from connected services like OneDrive.

Tools & Integrations

Select which tools and integrations your agent can access. You can choose up to three tools per agent.

Popular tool combinations include:

  • Tavily (web search)

  • Firecrawl (web scraping)

  • Python Code Interpreter (for document creation)

Note: While each agent is limited to three tools, you can create multiple agents and switch between them mid-conversation to access different tool sets.

System Prompt (Instructions)

The system prompt is the core instruction set that defines how your agent behaves. These instructions run behind the scenes every time you interact with your agent—so you only need to write them once.

Here are the 5 key components of a good system prompt:

  1. Role/Identity

    1. Clearly define who or what the AI should be (e.g., "You are a helpful customer service representative").

  2. Context

    1. Provide relevant background information, knowledge, or situational details the AI needs to respond appropriately.

  3. Instructions/Tasks

    1. Specify what the AI should do, how it should behave, and what actions it should take.

  4. Constraints/Guidelines

    1. Define boundaries, limitations, and rules for what the AI should avoid or how it should handle specific situations.

  5. Output Format

    1. Describe how responses should be structured, styled, or formatted (e.g., tone, length, use of markdown).

Creating a System Prompt

You can write system prompts manually, or use AI to help you create one:

  1. Open a second tab with an LLM (like Claude 4.5 Sonnet)

  2. Ask it to: "Create a user-friendly system prompt for an AI agent. Include core identity, behavioral rules, and instructions for how to generate outputs."

  3. Review the generated prompt

  4. Copy and paste it into your agent's Instructions field

  5. Customize as needed

Over time, you'll get better at writing these prompts and understanding what works best for your use cases.

Agent Icon (Optional)

Add a custom icon to help visually identify your agent. You can even use another Hatz tool to generate one!

Try this: Use Gemini 2.5 Flash Image (Nano Banana) in Hatz to prompt it to create an image icon for your agent.

Sharing and Deploying Agents

Once you've created your agent, you can control who has access to it.

Sharing Settings

By default, all new agents are set to Private (only you can access them).

Share with Organization Enable this to make the agent available to everyone in your organization. This is perfect for company-wide tools like brand voice agents or customer service assistants.

Share with Community Once an agent is shared with your organization, you can optionally share it with the broader Hatz community. Only do this if the agent doesn't contain company-specific information.

To update sharing settings:

  1. Click on your agent

  2. Select Share Settings

  3. Choose your sharing level

  4. Click Save Settings

Using Your Agent

You can interact with your agent in two ways:

From the Workshop

Interact with your agent directly from the Workshop view where you created it.

From Chat

  1. Navigate to the Chat interface

  2. Click the agent selector (look for the robot icon)

  3. Choose your agent from the list

  4. Start chatting!

The agent will automatically:

  • Follow the system prompts you configured

  • Reference uploaded sources

  • Use the tools and integrations you enabled

  • Maintain brand voice and guidelines

Example Interaction

Instead of having to say:

"You are a marketing agent. Here are our brand guidelines... Use our company voice... Format the output like this..."

You can simply say:

"Create a LinkedIn post about our newest features" + [link to changelog]

The agent already knows how to act because you configured all of that in the system prompt!

Tips for Building Effective Agents

Start Simple

Begin with a clear, specific use case. Don't try to build an agent that does everything—focus on one job done well.

Use Descriptive Names

Name your agent steps and sources clearly so you can easily identify them later.

Test Thoroughly

Run several test conversations with your agent to ensure it behaves as expected before sharing with your organization.

Leverage Multiple Agents

Since each agent can have up to three tools, consider creating specialized agents for different tasks:

  • Content Agent: OneDrive + Notion + Slack

  • Social Media Agent: LinkedIn + X + Reddit

  • Research Agent: Google Search + Tavily + Firecrawl

You can switch between agents mid-conversation to access different tool sets!

Keep System Prompts Updated

As your needs evolve, don't be afraid to update your agent's system prompts and sources.

Think About Your Team

What repetitive tasks take up your team's time? Those are perfect candidates for automation with agents.

Have ideas for agents you'd like to see? Let us know! We love hearing about creative use cases from the Hatz community.

Did this answer your question?