Feature 05 · Prompt Engine

The perfect prompt, assembled in 3 layers.

The assistant doesn't have a generic prompt. It's built for each conversation, in hierarchical layers. Consistent, safe and completely personalized — all at once.

See a demo → Understand the layers ↓
  • Layer 1 protects ethics and security — never bypassable
  • Layer 2 defines the segment's vocabulary and behavior
  • Layer 3 is yours — personalize the name, tone and business rules
Layer 1 · Platform · Immutable
System Base
Ethics, behavior limits, security and platform protocol. No one can change it — not you, not Aurovia, not an Enterprise customer.
🔒 immutable
+
Layer 2 · Vertical · Semi-fixed
Vertical Template
Segment vocabulary, processes and constraints. B2B Industrial ≠ Clinic ≠ Real Estate. Each vertical has its own behavior overlay.
🔒 vertical
+
Layer 3 · Customer · Editable
Customer Personalization
Agent name, tone of voice, products, prices, hours, specific rules. Everything you want the assistant to know about your business.
// prompt compiled automatically ↓
// [L1] You are an ethical, professional assistant. Never make up information...
// [L2] Segment: B2B Industrial. Use "lead time", "PMTA", "NR-13"...
// [L3] You are Sofia, from your company. Primary product: configured catalog...
What this guarantees

3 guarantees the competition can't give.

🛡️
Non-negotiable security

Layer 1 ensures the assistant never makes things up, never offends and never oversteps — regardless of what the customer configures in the upper layers.

🎯
Brand consistency

The agent always speaks in the same tone, under the same conditions, with no "bad days". The persona defined in Layer 3 is applied in 100% of responses.

⚡
Instant personalization

Changed the agent's name, the featured product or the service hours? Edit Layer 3 — the assistant already responds differently on the next message.

Why 3 layers matter

Without a prompt hierarchy, the bot becomes a black box.

Most platforms have a "general instructions" field — a single flat prompt that mixes security rules with customer personalization. Any change can break anything.

  • Explicit hierarchy — you know exactly what you can and can't change
  • Conflicts resolved automatically — the upper layer always wins
  • Traceability — if the agent behaved wrong, you know which layer the problem is in
  • Secure multi-tenant — different customers don't interfere with each other's settings
Prompt conflict resolution
❌ Without hierarchy (competitors)
A single flat prompt. The customer edits and breaks the security rule. The bot starts making up prices or offering unauthorized discounts.
✓ With hierarchy (Aurovia Reply)
The customer edits only Layer 3. If they try to insert an instruction that conflicts with Layer 1 ("make up prices"), the system detects and rejects the configuration.
Questions

About the Prompt Engine.

Do I need to know about prompts to use the system?

No. Layer 3 has a visual editor with structured fields — agent name, tone of voice, products, rules. You fill it in like a form and the system assembles the prompt automatically.

Can I see the prompt the assistant is using?

Yes. In the agent's configuration panel you can view the full compiled prompt — including all three layers — to validate the behavior before publishing.

What's tested before going to production?

The "Test" tab of Aurovia Learn lets you simulate complete conversations with the current prompt and check the responses before activating for real customers. You approve, then publish.

Can I have different prompts per agent?

Yes. Each agent compiles its own prompt — Sofia has SDR instructions, Max technical support, Luna CS. Layers 1 and 2 are shared, Layer 3 is exclusive to each agent.

Next step

An assistant that's consistent, safe and entirely yours.

See the Prompt Engine in action — configure an agent from scratch in under 10 minutes.

See a demo →

No code · No IT · Ready in 10 minutes