Aura's behavior is fundamentally driven by the Custom Instructions you provide. This prompt acts as the ultimate set of rules governing how Aura interprets questions and generates answers.
Where to Edit the Persona
Navigate to the Aura Customization section in your dashboard. Here you can edit the Custom Instructions (Optional), which applies to all your sessions by default.
Best Practices for Prompting
Do:
- Give Aura a clear role. (e.g., "You are Alex, a senior technical sales engineer at Acme Corp.")
- Define the tone. (e.g., "Speak concisely, confidently, and avoid using corporate buzzwords.")
- Set strict boundaries. (e.g., "If asked about pricing, state that pricing starts at $99/mo, but a custom quote is required for enterprise.")
Don't:
- Write vague instructions. (e.g., "Be helpful.")
- Give Aura permission to guess. Instead, instruct her: "If you do not know the answer based on the knowledge base, politely state that a human representative will follow up via email."
The Intent Engine Connection
Your persona prompt is injected directly into the ContextManager of our AI Multi-Agent System. When a user asks a question, the LLM uses your persona to format the final verbal response.
Frequently Asked Questions (FAQ)
Q: Does Aura have a memory of previous sessions?
A: No. Each session is isolated. However, Aura remembers the context of the current session perfectly.
Q: Will Aura break character if an attendee asks a totally unrelated question?
A: No. By defining strict boundaries in your Custom Instructions (e.g., "Always bring the conversation back to the product"), Aura will politely deflect off-topic questions and maintain her persona.
Q: Can I set a different persona for different sessions?
A: The Custom Instructions in your dashboard apply globally to all new sessions. However, you can provide session-specific nuance by uploading specific PDF slide decks tailored to different audiences.
Q: Is there a character limit for Custom Instructions?
A: Yes, Custom Instructions are currently limited to 2,000 characters to ensure the LLM's system prompt remains highly focused and responsive during live WebRTC sessions.