Knowledge Base

Create and manage knowledge bases so your agents have access to your organization-specific information

Overview

The Knowledge Base is where you store and organize information specific to your company, products, processes, or any content that your agents need to know. It’s like creating a “digital library” that your agents can consult to provide more accurate and up-to-date responses.

Why use Knowledge Base?

• 🎯 Specific responses: Your agents use information from your company, not just general knowledge
• 📚 Updated information: Keep data always current about products, policies, processes
• 🔍 Smart search: System automatically finds relevant information for each question
• 🏷️ Tag organization: Filter and organize content by specific categories

What are “chunks”? When you add a document, the system divides the content into small pieces called “chunks” (like pages of a book). This allows the agent to find exactly the relevant information without having to “read” the entire document for each question.

How to Create Your Knowledge Base

Step 1: Access Knowledge Base

Navigate to Knowledge

1. In the Evo AI main menu, click on “Knowledge”
2. You’ll see the list of all existing knowledge bases
3. Click on “Add Knowledge” to create a new base

You can create multiple knowledge bases to organize information by area: sales, support, products, etc.

Step 2: Choose Content Type

There are three ways to add information to your knowledge base:

1. Manual Entry (Text)

Ideal for: Policies, procedures, FAQs, specific information

1. Select “Manual Text Entry”
2. Name: Give it a descriptive title (e.g., “Return Policy”)
3. Content: Type or paste the text directly
4. Tags: Add keywords to organize (e.g., “sales”, “policy”, “returns”)

Practical example:

Name: Notebook Warranty Policy

Content: 
Our notebooks have a 12-month warranty for manufacturing defects.
The warranty covers: screen problems, keyboard, battery and internal components.
Does not cover: physical damage, drops, liquids or misuse.
To activate, customer must present invoice and complete registration.

Tags: warranty, notebook, policy, support

2. File Upload

Supported formats:

📄 Documents:

• PDF, DOCX, XLSX, PPTX, HTML

📝 Text:

• TXT, MD, MDX

🖼️ Images:

• PNG, JPG, JPEG, TIFF, BMP

Maximum limit: 50MB per file

How to do it:

1. Select “File Upload”
2. Drag and drop the file or click to select
3. Name: The system suggests automatically, but you can change it
4. Wait for processing - the system extracts text and creates chunks automatically
5. Review the automatically generated tags

Tip: For large PDFs, the system divides into smaller chunks. For example, a 50-page manual becomes about 20-30 thematic chunks.

3. URL Processing

Ideal for: Websites, online documentation, web articles

Features:

• Simple URL: Processes a specific page
• Automatic sub-pages: Discovers and processes linked pages from the same domain

How to configure:

1. Select “URL Processing”
2. URL: Paste the link to the page or online document
3. Include Sub-pages: ✅ Check to process related pages
4. Maximum sub-pages: Set limit (recommended: 10-50 depending on site)
5. Wait for processing - may take a few minutes for large sites

Practical example:

URL: https://company.com/documentation/products
Include Sub-pages: ✅ Yes
Max Sub-pages: 25

Result: System processes main page + up to 25 related pages
automatically (e.g., /products/notebook, /products/smartphone, etc.)

Advantages:

• Automatic web content updates
• Captures complete documentation structure
• Saves time vs. manual upload

Step 3: Automatic Tags and Metadata

Automatic Tag System

How it works:

• The system analyzes content using AI
• Automatically generates tags based on main themes
• You can edit, add or remove tags as needed
• Tags help with search and filters by agents

Examples of automatic tags:

Document: "iPhone 15 Manual"
Generated tags: smartphone, apple, iphone, manual, setup, usage

Document: "HR Policy - Vacation"  
Generated tags: hr, vacation, policy, employees, rights

Document: "Q3 Financial Report"
Generated tags: financial, report, Q3, revenue, expenses, profit

How to customize:

• ✅ Keep relevant tags that facilitate search
• ➕ Add specific tags from your business
• ❌ Remove generic tags that don’t add value
• 🎯 Use consistent tags between similar documents

Chunk Metadata

What is metadata?

• Extra information about each chunk (piece of content)
• Includes: source, date, file type, size, relevance
• Helps the system find more precise information

Automatically generated metadata:

Chunk 1 - iPhone Manual (Pages 1-2):
- Source: manual_iphone_15.pdf
- Size: 1.2kb
- Themes: initial setup, first use
- Relevance: high for "how to configure"

Chunk 2 - iPhone Manual (Pages 10-12):
- Source: manual_iphone_15.pdf  
- Size: 0.8kb
- Themes: camera, photos, videos
- Relevance: high for "taking photos"

How the system uses it:

• Contextual search: Finds most relevant chunk for each question
• Smart filtering: Avoids unnecessary information
• Ranking: Prioritizes most useful chunks for the answer

Testing Your Knowledge Base

Search and Verification

How to test if it's working

Search Tab:

1. After creating content, go to the “Search” tab in Knowledge Base
2. Type a question related to the content you added
3. See the results: System shows most relevant chunks
4. Check relevance: Do the results make sense for your question?

Test example:

Added content: Notebook warranty manual

Test question: "How long does the notebook warranty last?"

Expected result: Chunk with information "12 months warranty"

If it doesn't appear: Check if tags include "warranty" and "notebook"

Improving search quality

If search isn’t good:

1. Refine tags:

• Add synonyms: “warranty” + “coverage” + “protection”
• Use terms that customers actually say
• Avoid overly technical tags

2. Improve content:

• Use clear and direct language
• Include common questions in the text
• Add sufficient context in each chunk

3. Test different questions:

• Formal question: “What is the warranty period?”
• Casual question: “How long does the notebook have warranty?”
• Specific question: “Notebook with defect, does warranty cover?”

4. Adjust large chunks:

• If chunk is too long, divide into smaller themes
• If too small, may lack context
• Ideal: 200-500 words per chunk

Using Knowledge Base with LLM Agents

Agent Configuration

Activating Load Knowledge

How to connect Knowledge Base to your agent:

1. Open agent settings of your LLM agent
2. Agent Settings section: Find the “Load Knowledge” option
3. ✅ Activate Load Knowledge: Check the checkbox
4. Configure tag filters (optional): Select specific tags
5. Save settings

What happens when active:

• Agent automatically receives a tool called load_knowledge
• Can search for information in Knowledge Base when needed
• Uses tags to filter relevant content
• Integrates specific knowledge into responses

Tag configuration:

Sales Agent:
Selected tags: sales, products, prices, warranty
→ Accesses only sales-related content

Support Agent:  
Selected tags: support, problems, solutions, tutorials
→ Accesses only technical support content

General Agent:
No tags specified
→ Accesses entire Knowledge Base

How it works in practice

Automatic flow:

1. Customer asks question: “What’s the warranty on notebook X?”
2. Agent analyzes: Identifies need for specific information
3. Automatic search: Uses load_knowledge tool to find answer
4. Filter by tags: If configured, searches only in relevant content
5. Integrates in response: Combines found knowledge with LLM reasoning
6. Final response: Customer receives precise and updated information

Practical example:

Customer: "My notebook broke, do I have warranty?"

Agent:
1. 🔍 Automatic search in Knowledge Base
2. 📄 Finds: "12 months warranty for manufacturing defects"
3. 🤖 Processes with LLM: Analyzes situation + policy
4. 💬 Response: "Yes! Your notebooks have 12 months warranty for 
   manufacturing defects. To use the warranty, you'll need the 
   invoice and registration form completion. What kind of problem 
   is happening?"

Advantages:

• ✅ Always updated information
• ✅ Consistency between agents
• ✅ Responses based on real data
• ✅ Reduces AI “hallucinations”

Tag filter strategies

Why use filters:

• Avoids irrelevant information in responses
• Improves search speed
• Reduces confusion between different areas
• Allows agent specialization

Recommended strategies:

1. By department:

Sales Agent: sales, products, prices, promotions
HR Agent: hr, employees, policies, benefits  
IT Agent: it, systems, passwords, technical-problems

2. By customer type:

B2B Agent: companies, corporate, volumes, contracts
B2C Agent: consumer, retail, individual, simple

3. By product:

Notebook Agent: notebook, laptop, computer, hardware
Smartphone Agent: mobile, smartphone, mobile, apps

4. No filter (general agent):

Supervisor Agent: No tags - full access
Triage Agent: No tags - can direct to specialists

Organizing Your Knowledge Base

Best Practices

Organizational structure

How to organize multiple knowledge bases:

By business area:

📚 Sales KB
- Product catalog
- Pricing policies
- Sales scripts
- Common objections

📚 Support KB  
- Technical manuals
- Troubleshooting
- FAQs
- Procedures

📚 HR KB
- Internal policies
- Benefits
- Procedures
- Company culture

By content type:

📚 Official Documents KB
- Standard contracts
- Corporate policies
- Legal procedures

📚 Practical Knowledge KB
- Tutorials
- Success cases
- Tips and tricks

Specialized agents access specific KBs for maximum efficiency.

Keeping content updated

Maintenance strategies:

1. Periodic review:

• Set schedule (monthly/quarterly)
• Check outdated information
• Update prices, policies, products
• Remove obsolete content

2. Version control:

• Use descriptive names with dates
• “Product_Catalog_2024_Q1”
• “Warranty_Policy_v2_January2024”

3. Automatic URLs:

• For frequently changing sites
• Configure automatic reprocessing
• Monitor if URLs are still active

4. Area managers:

• Sales: Updates products and prices
• HR: Updates internal policies
• Support: Updates technical procedures

Effective tag system

Tag standardization:

Use consistent categories:

By area: sales, support, hr, financial, marketing
By product: notebook, smartphone, tablet, accessory
By type: policy, manual, tutorial, faq, procedure
By urgency: critical, important, normal, informative

Recommended conventions:

• Lowercase: sales, not Sales
• Hyphen for compounds: technical-problem, not technical problem
• Consistency: always “smartphone”, never mix with “mobile”
• Hierarchy: product > product-notebook > product-notebook-gaming

Practical examples:

Document: "How to replace iPhone screen"
Tags: support, smartphone, iphone, repair, screen, tutorial

Document: "Gaming notebook price table"  
Tags: sales, notebook, gaming, prices, table, 2024

Common Use Cases

E-commerce

Knowledge Base for online store:

• Complete product catalog
• Return/exchange policies
• Size guides
• Delivery FAQs

Benefited agents:

• Sales: Precise recommendations
• Support: Quick resolution
• After-sales: Updated policies

SaaS/Software

Knowledge Base for software:

• Technical documentation
• Usage tutorials
• Common troubleshooting
• Changelog and updates

Benefited agents:

• Support: Technical solutions
• Onboarding: Setup guides
• Sales: Features and benefits

Services

Knowledge Base for consulting:

• Proprietary methodologies
• Success cases
• Standard proposals
• Expert profiles

Benefited agents:

• Sales: Personalized proposals
• Delivery: Consistent execution
• Relationship: Complete history

Internal Support

Corporate Knowledge Base:

• HR policies
• Internal procedures
• Systems and passwords
• Culture and values

Benefited agents:

• HR: Standardized responses
• IT: Automated support
• Management: Consistent information

Troubleshooting

Common problems

1. Agent doesn’t find information:

• ✅ Check if Load Knowledge is active
• ✅ Confirm tags are configured correctly
• ✅ Test manual search in Search tab
• ✅ Add synonyms in tags

2. Too generic responses:

• ✅ Improve KB content quality
• ✅ Add more context in documents
• ✅ Use smaller and more specific chunks
• ✅ Refine tags for greater precision

3. Upload failed:

• ✅ Check file size (max 50MB)
• ✅ Confirm supported format
• ✅ Test smaller file first
• ✅ Convert to PDF if necessary

4. URL doesn’t process:

• ✅ Check if URL is publicly accessible
• ✅ Test URL in browser first
• ✅ Reduce number of sub-pages
• ✅ Wait longer for large sites

5. Search returns irrelevant results:

• ✅ Refine tags to be more specific
• ✅ Divide long documents into smaller chunks
• ✅ Use clearer language in content
• ✅ Test different ways to ask the question

Next Steps

Configure LLM Agent

Learn how to configure agents to use your Knowledge Base

Agent Settings

Understand all Load Knowledge options

Specialized Agents

Create specialized agents by knowledge area

Best Practices

Optimize performance and response quality

---

🎓 Congratulations! Now you know how to create and manage knowledge bases to make your agents smarter and more precise with your organization-specific information!