Building Your Agent's Knowledge Base: The Cortex Onboarding Experience

When you sign up for most AI platforms, you write a system prompt. You try to cram your company's entire context, your team's procedures, your agent's personality, and your business logic into a few hundred words of compressed technical instructions.

This approach has an obvious problem: system prompts are hard to write, hard to maintain, and invisible to everyone except the person who wrote them. They're source code, not documentation.

Cortex takes a different approach. Your agent's knowledge comes from a structured, searchable knowledge base built through a guided interview. This knowledge base is documentation: readable, maintainable, versionable, and accessible to your whole team.

The Five Sections of the Knowledge Hub Builder

When you create a Cortex agent, you walk through the Knowledge Hub Builder: an AI-guided interview with five sections. Each section captures different kinds of knowledge your agent needs.

Section 1: Company Overview

This section establishes context: What does your organization do? What problems do you solve? What's your market position?

The AI interviewer digs deeper: What are your core values? Who are your customers? What differentiates you from competitors? What industries do you serve?

The answers become your agent's understanding of your organization's mission and identity. This isn't used in a system prompt; it's stored as structured documentation that your agent can reference.

Example answers might look like:

"Acme Software provides project management solutions for distributed teams. We focus on asynchronous workflows and remote-first organizations. Our customers are typically 20-500 person companies in tech, marketing, and professional services. We compete on ease of use and integration ecosystem, not on feature breadth."

This becomes documentation your agent understands and can apply when helping customers, answering questions about your product, or representing your company.

Section 2: Team Structure

How is your team organized? What roles exist? Who owns which functions?

For a customer success team, this might capture: "Sarah leads the team. The team has three customer success managers (each with regional responsibility), one technical onboarding specialist, and one operations coordinator. Sarah reports to the VP of Customer Success."

For a product team, it might be: "The product team has three product managers (backend, frontend, platform), one product operations manager, and one designer. The team reports to the VP of Product."

This knowledge helps your agent understand who to escalate to, what expertise exists on your team, and how decisions get made.

Section 3: Tools and Tech Stack

What systems does your team actually use? What matters for the agent's effectiveness?

This might include CRM tools, project management platforms, internal wikis, communication channels, monitoring systems, or specialized software. For a support agent, this might include your ticketing system, knowledge base, and customer communication platform. For a product team, this might include design tools, code repositories, and analytics platforms.

The agent understands what systems are available, what integrations are possible, and where information lives. This prevents the agent from suggesting tools that don't exist in your environment.

Section 4: Goals and Priorities

What should this specific agent focus on? What are the success metrics?

For a customer support agent, goals might be: "Resolve customer issues in under 2 hours. Prioritize data access issues and billing questions. Escalate custom integration work to the technical team. Maintain a friendly, patient tone regardless of customer frustration level."

For a sales enablement agent, goals might be: "Prepare meeting agendas and talking points. Research customer backgrounds before calls. Suggest relevant case studies and testimonials. Flag customer pain points that suggest upsell opportunities."

Goals give the agent direction. They help it prioritize what matters and what doesn't.

Section 5: Agent Persona and Rules

How should this agent behave? What personality should it adopt? What are the hard boundaries?

This captures behavioral guidance: "You're direct and practical. You get to the point quickly. When customers seem frustrated, you acknowledge their frustration and offer concrete next steps. You never make promises about features not yet released. You never access customer data beyond what's necessary for support. When you don't know something, you say so immediately."

This is where you define the agent's voice, constraints, and decision-making principles.

From Answers to Knowledge Base

Here's what makes Cortex different from a system prompt: your answers are processed into structured markdown documents.

These are not hidden instructions buried in model configuration. They're explicit documentation:

# Acme Software Company Overview

## Mission and Products
Acme Software provides project management solutions for distributed teams...

## Target Market
20-500 person companies in tech, marketing, and professional services...

## Competitive Positioning
We compete on ease of use and integration ecosystem...

---

# Team Structure

## Leadership
Sarah leads the customer success team, reporting to the VP of Customer Success...

## Roles and Responsibilities
- **Customer Success Managers**: Regional responsibility for customer health...
- **Technical Onboarding Specialist**: Handles initial customer setup...
- **Operations Coordinator**: Manages scheduling and logistics...

---

# Tools and Tech Stack

## Systems in Use
- **CRM**: Salesforce
- **Ticketing**: Zendesk
- **Knowledge Base**: Confluence
- **Communication**: Slack

---

# Goals and Priorities

## Primary Goals
1. Resolve customer issues in under 2 hours
2. Prioritize data access issues and billing questions
3. Escalate custom integration work to the technical team

## Behavioral Guidelines
- Maintain friendly, patient tone regardless of customer frustration
- Acknowledge frustration and offer concrete next steps
- Never make promises about unreleased features

---

# Agent Persona and Rules

## Voice and Style
You're direct and practical. You get to the point quickly.

## Constraints
- Never access customer data beyond what's necessary for support
- If you don't know something, say so immediately
- Never override the technical team's decisions

This is your knowledge base: readable, maintainable, versionable. You can update it without needing a developer. You can share it with your team. You can see what the agent knows.

System Prompt vs. Knowledge Base

This distinction matters more than it might seem.

A system prompt is:

• Hidden from most users
• Hard to update without redeploying
• Binary: either in the prompt or out
• Difficult to organize for large knowledge sets
• Not easily searchable

A knowledge base is:

• Explicit and transparent
• Easy to update and version control
• Granular: you add and modify specific sections
• Naturally organized and searchable
• Your whole team can understand and maintain it

When your agent references its knowledge base, it's performing semantic search on actual documentation. When your agent needs to check the rules, it's reading the rules you wrote. This creates auditability. You can see what knowledge your agent is accessing. You can trace decisions back to documented principles.

Dynamic Learning Through Scopes

The knowledge base you build is your agent's initial knowledge. But Cortex's three-scope memory system means your knowledge base grows over time.

Agent scope captures what this specific agent learns from interactions. Team scope captures knowledge shared across your team. Company scope captures organization-wide policies and glossary.

Your Knowledge Hub Builder knowledge forms the foundation at company and team scopes. As the agent operates, agent-level memory accumulates. As multiple agents discover patterns, knowledge graduates to higher scopes.

The knowledge base is alive. It's not static documentation written once and forgotten; it's the foundation for a learning system.

Getting Started

The Knowledge Hub Builder interview typically takes 3-4 minutes. You answer questions, the AI interviewer digs deeper, and by the end you have a structured knowledge base ready for your agent.

This is fundamentally different from the traditional approach of writing system prompts. You're not trying to compress organizational knowledge into technical prose. You're documenting your organization, your team, your goals, and your agent's role in natural language.

Then Cortex indexes that documentation, makes it searchable, and gives your agent the ability to reference it accurately.

That's how you build agent knowledge bases that scale. Not hidden system prompts, but explicit, maintainable, team-accessible documentation that your agent understands and your team can verify.

Ready to build a knowledge base for your AI agent? Sign up at launchcortex.ai and try the Knowledge Hub Builder in minutes.