Knowledge base
Your agent needs information to answer customer questions effectively. Stellar provides three different ways to give your agent knowledge, each suited for different types of information and use cases.
Options for providing knowledgeβ
1. General knowledge baseβ
What it is: A central repository of information available to your agent throughout all conversation states.
When to use it:
- Static information that doesn't change frequently
- General company information, policies, and procedures
- Product features, pricing, and specifications
- FAQs and common customer questions
- Information needed across multiple conversation scenarios
How to add it: Navigate to the Knowledge tab in your agent editor and enter text in the Source Information field.
Example:
# Product Information
Our main product is CloudSync, a file synchronization service for businesses.
## Features
- Real-time file synchronization across devices
- Team collaboration with shared folders
- 256-bit AES encryption for all files
- 30-day file recovery
- Unlimited storage for Enterprise plans
## Pricing
- Starter: $9/month for 100GB
- Business: $25/month for 1TB
- Enterprise: Custom pricing with unlimited storage
## Support Hours
- Email support: 24/7
- Phone support: Monday-Friday, 9am-6pm EST
- Enterprise customers receive 24/7 phone support
## Common Questions
Q: Can I try CloudSync before buying?
A: Yes, we offer a 14-day free trial with full access to Business plan features.
Q: How do I cancel my subscription?
A: You can cancel anytime from your account settings under Billing > Manage Subscription.
Q: Is my data secure?
A: All files are encrypted with 256-bit AES encryption both in transit and at rest. We are SOC 2 Type II certified.
To ensure your agent consistently references the knowledge base, explicitly mention it in your agent's personality instructions. For example: "Always consult your source information before answering customer questions" or "Only provide information that is explicitly stated in your knowledge base."
2. State-specific instructionsβ
What it is: Knowledge and instructions added directly to specific conversation states in the Dialogue Builder.
When to use it:
- Context-specific information only relevant in certain situations
- Specialized instructions for handling specific scenarios
- Information that would clutter the general knowledge base
- Details that only matter during particular conversation phases
How to add it: In the Dialogue Builder, edit a state and include the relevant information in the Instructions field.
Example for a "Billing Support" state:
You are helping the customer with billing questions.
## Payment Methods
- Credit card (Visa, Mastercard, Amex)
- PayPal
- Bank transfer (Enterprise customers only)
## Refund Policy
- Full refund within 30 days of purchase, no questions asked
- After 30 days: Prorated refund based on remaining subscription period
- Refunds processed within 5-7 business days
## Common Billing Issues
- Failed payments: Usually due to expired card or insufficient funds
- Duplicate charges: Rare, but immediately refundable
- Upgrade/downgrade: Takes effect on next billing cycle
This information is only active when the agent is in the Billing Support state, keeping other states focused on their specific purposes.
3. Dynamic knowledge via actionsβ
What it is: Real-time data retrieved from your external systems through API integrations.
When to use it:
- Information that changes frequently or in real-time
- Customer-specific data (order status, account details, bookings)
- Personalized information that requires authentication
- Inventory, availability, or pricing that updates regularly
- Data stored in your existing systems
How it works: Configure Actions that connect to your APIs. The agent calls these actions during conversations to fetch current information.
Example use cases:
Order Status Lookup:
- Customer: "Where is my order?"
- Agent calls
get_order_statusaction with order number - Returns real-time shipping information
Account Balance:
- Customer: "What's my current balance?"
- Agent calls
get_account_balanceaction - Returns up-to-date account information
Appointment Availability:
- Customer: "Do you have any appointments available tomorrow?"
- Agent calls
check_availabilityaction - Returns current open slots
See the Actions documentation for details on configuring API integrations.
Choosing the right approachβ
Consider these key factors when deciding where to place information:
Context length and focusβ
The challenge: AI models have limited context windows, and too much information can dilute important instructions or overwhelm the agent.
General knowledge base:
- β Use for essential information needed across multiple states
- β Keep it conciseβonly include what's frequently referenced
- β Avoid large policy documents or extensive product catalogs
- β Don't include information only relevant to specific scenarios
State-specific instructions:
- β Use for detailed information only needed in specific contexts
- β Good for lengthy policies or procedures relevant to that state
- β Keeps other states focused and uncluttered
- β οΈ Information is only available when the agent is in that specific state
Dynamic via actions:
- β Perfect for very large documents that can be retrieved on-demand
- β Retrieves only what's needed, when it's needed
- β No impact on context length until actually fetched
- β οΈ Requires API development and maintenance
Example: A 50-page return policy document would overwhelm the general knowledge base. Instead, put a summary in the general knowledge base and either: (1) include the full policy in a "Returns & Refunds" state, or (2) create an action to fetch specific sections on-demand.
Frequency of useβ
How often will this information be needed?
General knowledge base:
- β Information needed in most conversations
- β Commonly asked questions
- β Core product/service details
State-specific instructions:
- β Information only relevant in specific situations
- β Specialized knowledge for particular conversation phases
- β Don't duplicate general knowledge here
Dynamic via actions:
- β Information needed occasionally
- β Deep details that may or may not be required
- β οΈ API call overheadβonly use when worth the retrieval time
Rate of changeβ
How frequently does this information update?
General knowledge base:
- β Stable information (policies, company info, basic features)
- β οΈ Requires manual updates when information changes
- β Not suitable for frequently changing data
State-specific instructions:
- β Information that changes occasionally
- β οΈ Requires updating each relevant state when changed
- β Not suitable for real-time data
Dynamic via actions:
- β Information that changes frequently
- β Real-time or near-real-time data
- β Updates automatically when your system updates
- β No need to update Stellar when data changes
Personalization needsβ
Is this information the same for everyone or customer-specific?
General knowledge base:
- β Same information for all customers
- β Cannot be personalized
State-specific instructions:
- β Same information for all customers in that context
- β Cannot be personalized
Dynamic via actions:
- β Customer-specific information
- β Requires authentication/identification
- β Can return different data per customer
Next stepsβ
Once you've configured your knowledge:
- Test your agent in the Playground to ensure it uses knowledge correctly
- Build dialogue flows to handle different conversation types
- Set up actions for dynamic data retrieval
- Configure guardrails to keep conversations safe and on-topic