Prompts
Each assistant has several configurable prompts that control how it behaves in different contexts. You can customize these on the assistant detail page under the Prompts tab.
Personality Prompt
The personality prompt is set when you create or edit the assistant. It defines who the assistant is — its role, expertise, tone, and behavior. This is the primary instruction set that shapes every interaction.
You can include template tags in your personality prompt to dynamically insert context:
| Tag | Description |
|---|---|
{assistant_name} | The assistant’s name |
{user_name} | The current user’s name |
{user_role} | The current user’s role |
{user_bio} | The current user’s bio |
{tools} | Descriptions of available tools |
{skills} | Descriptions of enabled skills |
{knowledge} | Core knowledge sources |
{guidelines} | Content guidelines from the project |
{agents} | Other available assistants for consultation |
{project} | Project context information |
{deliverables} | Recent deliverables in the conversation |
{mode} | Mode-specific instructions |
Any section tags you don’t explicitly include will be automatically appended to the end of your prompt, so the assistant always has access to all relevant context.
System Instructions
The system instructions field lets you override the entire system prompt sent to the LLM. When left blank, the system prompt is assembled automatically from the personality prompt, tool instructions, user context, and other dynamic sections. When you write custom system instructions, they completely replace this default assembly.
You can preview the default system prompt on the assistant detail page under the Prompts tab. This shows the full prompt template including the personality prompt content, template tags for dynamic sections, and the user context block. You can copy it, modify it, and paste it back to take full control.
Template tags like {tools}, {skills}, {knowledge}, and {user_name} work in the system instructions field. Unlike the personality prompt, missing section tags are not auto-appended — you control exactly what’s included. This means you can remove sections you don’t need or rearrange the prompt structure.
Leave blank to use the default assembly, or write your own to fully replace it.
Voice Instructions
These instructions apply specifically when the assistant is in a voice conversation via WebRTC. The defaults guide the assistant to:
- Keep responses concise and conversational
- Avoid markdown formatting (since responses are spoken)
- Tell the user before executing a tool (“Let me look that up…”)
- Use the canvas for visual information via the
showtool
Customize these if you need different voice behavior.
Chat Instructions
These instructions apply during text chat sessions. The defaults explain the split-screen canvas interface and guide the assistant to:
- Use the
showtool to display visual cards on the canvas - Keep chat responses brief when card content is displayed
- Number cards for easy user reference
Customize these to change how the assistant uses the chat interface.
Resetting Prompts
Each prompt type has a Reset button that clears your custom instructions and reverts to the defaults. This only affects the specific prompt you reset — other customizations are preserved.