How AI Thinks

Your AI agent processes every conversation through three stages: it loads its instructions, searches for relevant information, and then uses tools to respond. Understanding these stages helps you configure your agent more effectively.

Stage 1: System Prompt

Before your AI agent responds to a single customer message, it receives a briefing — similar to a new employee reading the company handbook before their first shift. This briefing is called the system prompt.

Persona

The persona defines how your agent presents itself: its name, tone, and communication style. A formal B2B assistant reads very differently from a casual consumer chatbot, and both are configured here.

Rules

Rules are specific behavioral instructions. Examples:

• "If a customer asks about returns, always mention the 30-day policy."
• "Never discuss competitor pricing."
• "When recommending a product for back pain, always suggest the ErgoSeat Pro first."

Rules shape how the agent handles specific situations — they are the guardrails that keep responses aligned with your business needs.

Context

Context includes background information about your business: your brand, your products, your support policies. The agent uses this as a foundation for every response.

Stage 2: Vector Search

The AI does not memorize your entire catalog and knowledge base. Instead, when a customer asks a question, the agent searches for the most relevant information at that moment.

How Search Works

Your data is converted into mathematical representations called vectors. When a question arrives, the agent converts that question into a vector as well, then finds the stored data that is mathematically closest — meaning most semantically similar — to what was asked.

This is why the search understands meaning, not just keywords. "My back hurts" and "lumbar pain" point to the same results.

What Gets Searched

• Products — your product catalog
• Knowledge base — articles, FAQs, guides you have added
• Rules — relevant behavioral rules for the current situation

Why This Matters

Because the agent retrieves only what is relevant for each conversation, it stays accurate and focused. It does not confuse products from different categories or give generic answers when specific information exists.

Stage 3: Tools

Tools extend what the AI can do beyond generating text. Instead of just describing a product, the agent can display it. Instead of telling a customer to check their email, the agent can send one.

Built-in Tools

Every agent comes with core tools out of the box:

• Product search — looks up products based on the conversation
• Product display — shows a formatted product card directly in the chat

Plugin Tools

Additional capabilities can be enabled through plugins:

• Email service — sends transactional emails (order confirmations, follow-ups)
• Custom APIs — connects your agent to external systems

How Tools Work

The agent decides when to use a tool based on the conversation. A customer saying "show me something for headaches" triggers the product search tool automatically. The agent does not need to be explicitly told to search — it infers the intent and acts.

Putting It All Together

Every response follows the same sequence:

1. System Prompt — sets the personality, rules, and business context
2. Vector Search — retrieves the information relevant to this specific question
3. Tools — take action (search, display, email) based on what was retrieved

This is why your agent can feel knowledgeable and consistent at scale — not because it memorizes everything, but because it assembles the right information and capabilities for each conversation.