Overview
Writing effective agent instructions is critical for agent performance, but it can be challenging. The Build Assistant removes that friction by generating complete instruction sets, suggesting improvements, and adding personality—all tailored to your agent's name, description, and purpose.
Who This Is For
The Build Assistant is available to all Hatz AI users creating or editing agents in the Workshop editor. It's especially useful for:
Users building their first agent who need a strong starting point
Teams refining existing agents to improve clarity and consistency
Anyone adding tone and personality to make agents feel more natural
Where to Find It
The Build Assistant appears on the right side of the agent editor when you open any agent in the Workshop.
Wide screens: Always visible in the right panel
Smaller screens: Access via the "Build Assistant" tab at the top of the right panel
At the top, you'll see a live preview card showing your agent's icon, name, and description as you fill them in.
How It Works
Greeting Message Preview
Just below the agent preview card, the Build Assistant shows how your agent will open conversations.
Default greeting (if no custom greeting set): "Hello! I'm [your agent's name]. How can I help you today?"
Custom greeting: Updates in real time as you type in the agent configuration form on the left
Quick Actions
When you first open the Build Assistant, three Quick Action buttons appear. Click one to get started instantly—no typing required.
Help me write instructions
Generates a complete set of agent instructions based on your agent's name, description, and categories. Produces a thorough system prompt with behaviors, tone, and constraints appropriate for your agent's purpose.
If instructions already exist, this button changes to Rewrite and improve and enhances the existing content rather than starting over.
Suggest improvements
Reviews your current instructions and produces a revised version with targeted improvements focused on clarity, completeness, edge cases, and best practices.
If no instructions exist yet, this suggests what a strong instruction set should include.
Add tone & personality
Layers personality traits, communication style, and tone guidelines into your existing instructions. Makes the agent feel more natural and engaging while staying professional.
If no instructions exist yet, writes personality and tone guidelines from scratch based on your agent's name and description.
The Chat Input
Below the message area is a free-form text input where you can ask the Build Assistant anything about your agent's instructions.
Examples of requests:
"Make the tone more formal"
"Add a section about how to handle questions the agent doesn't know the answer to"
"Rewrite this to be shorter and more direct"
"Add instructions telling the agent to always respond in bullet points"
"Make it sound friendlier"
How to use it:
Press Enter to send your message
Press Shift + Enter to add a line break without sending
The assistant remembers the last 20 messages in your conversation, so you can iterate without repeating context
How Responses Work
Streaming Responses
The Build Assistant streams responses in real time—you'll see text appear word by word as it's generated.
A spinning indicator appears on the Send button while the assistant is working
The input is disabled until the response completes to prevent accidental interruptions
Thinking Indicator
For complex requests, you may see a pulsing three-dot indicator before text appears. This means the assistant is processing your request.
Apply to Instructions
Every response includes an "Apply to Instructions" button at the bottom of the message.
Clicking this copies the assistant's response directly into the Instructions field on the left panel, replacing the current content
Nothing changes until you choose to apply it—you always review first
Once applied, the button disappears and the assistant confirms: "Applied to your instructions field. You can edit it further on the left, or ask me for more refinements."
You can apply any previous response in the conversation, not just the most recent one
Context Awareness
The Build Assistant automatically knows:
Agent name – from the Name field
Description – what the agent is for
Categories – any categories you've assigned
Current instructions – whatever is in the Instructions field when you send a message
If you update the agent name or description during your session, those changes are picked up automatically the next time you send a message.
Tips for Best Results
Fill in the name and description first.
The Build Assistant uses these as its starting point. Even a short, clear description dramatically improves output quality.
Use quick actions to get a strong first draft, then refine.
Quick actions produce solid, complete starting points. Once you have something in the instructions field, switch to the chat input for specific adjustments.
Be specific in follow-up messages.
Instead of "make it better," try "make the tone less formal" or "add more detail about how the agent should handle complaints." The more specific you are, the more useful the response.
Apply and then keep iterating.
You don't have to get it perfect in one go. Apply a suggestion, review it, and ask the Build Assistant for another pass if needed. The conversation history stays intact.
The instructions field is always editable.
Anything the Build Assistant writes is a starting point. After applying a suggestion, you can edit the instructions directly on the left side of the editor, then come back and ask the assistant to improve your changes.
Limitations
Tool configuration is not included.
The Build Assistant does not currently know about the tools you've connected to the agent. Tool configuration is handled separately in the tools section of the agent editor. You'll need to manually add any tool-specific guidance to your instructions.
Conversation history resets between sessions.
The conversation resets each time you navigate away from the agent editor and come back. Within a single editing session, the full conversation history (up to 20 messages) is preserved.
Apply replaces existing instructions.
Clicking "Apply to Instructions" replaces the entire contents of the Instructions field. There is no automatic undo. To revert, you must paste the previous instructions back in manually or ask the Build Assistant to regenerate the version you preferred.
AI model dependency.
The Build Assistant uses the AI model you've selected for the agent in the configuration form. If you haven't chosen a model yet, it uses the platform default. Response quality and style may vary depending on the selected model.
Greeting Message character limit.
The Greeting Message field has a 250 character limit. The character count is shown below the field and turns red if you're approaching the limit.
Quick Action behavior changes with context.
The Quick Action buttons adapt based on whether instructions already exist. For example, "Help me write instructions" becomes "Rewrite and improve" if instructions are present. This is expected behavior, not a bug.
Related Features
Greeting Message
Set the opening message your agent sends at the start of every conversation. Found in the agent configuration form on the left side of the editor. The Build Assistant preview updates live as you type it.
Personalization Toggle
Controls whether the agent adapts its responses based on individual user preferences. Found in the agent configuration form. Can be set to inherit the account default, always on, or always off per agent.
Per-Agent Tool Configuration
Some tools (like Outlook) can be scoped to specific folders or channels per agent. Configured separately in the tools section of the agent editor.