Skip to main content

Knowledge Base

Store and retrieve knowledge for AI-powered features. Knowledge bases store information that AI can search and reference.

What is a Knowledge Base?​

A knowledge base is a vector database that stores information as vector embeddings, enabling:

  • Semantic search
  • AI context retrieval
  • FAQ matching
  • Document Q&A

It can contain:

  • Documents and text
  • FAQs and answers
  • Product information
  • Any searchable content

AI nodes can search your KB to provide accurate, contextual answers.

Knowledge Base Types​

CRM Knowledge Bases​

Auto-synced from CRM data:

  • Contacts
  • Companies
  • Deals
  • Notes
  • Tasks

Custom Knowledge Bases​

Upload your own content:

  • Documents (PDF, Word)
  • Web pages
  • Text files
  • FAQs

Creating a Knowledge Base​

  1. Navigate to Automation → Knowledge Base
  2. Click + Create Knowledge Base (or + New)
  3. Enter:
    • Name - Descriptive name
    • Description - What it contains
  4. Click Create

From Documents​

  1. Go to Automation → Knowledge Base
  2. Click + New
  3. Upload files or enter URLs
  4. Configure chunking settings
  5. Process and embed

Automatic Sync​

CRM knowledge bases sync automatically:

  • New records added
  • Updates reflected
  • Deletions removed

Adding Content​

Manual Entry​

  1. Open your knowledge base
  2. Click + Add Entry
  3. Enter:
    • Title - Entry title
    • Content - Main text
    • Metadata - Tags, categories
  4. Save

File Upload​

Supported formats:

  • PDF documents
  • Word documents (.docx)
  • Text files (.txt)
  • Markdown (.md)
  1. Click Upload Files
  2. Select files
  3. Files are parsed and chunked automatically

URL Import​

Import content from websites:

  1. Click Import from URL
  2. Enter URL
  3. Content is scraped and added

API Import​

curl -X POST https://api.expedify.ai/api/knowledge-base/{kb_id}/entries \
  -H "Authorization: Bearer API_KEY" \
  -d '{
    "title": "Product FAQ",
    "content": "Your product information here..."
  }'

Content Chunking​

Large documents are automatically split into chunks for better search:

SettingDescription
Chunk sizeCharacters per chunk (default: 1000)
OverlapCharacters shared between chunks
StrategyBy paragraph, sentence, or fixed

Using in Workflows​

KB Search Node​

Add a Knowledge Base Search node:

  1. Select knowledge base
  2. Configure:
    • Query - What to search for (from variable or static)
    • Top K - Number of results
    • Threshold - Minimum relevance score
  3. Use results in downstream nodes

With LLM Nodes​

Combine KB search with LLM:

[KB Search] → [LLM Node]
              Prompt: "Using this context: {{kbsearch_1.results}},
                      answer: {{user_question}}"

Example Use Cases​

  • Customer support: Search past tickets
  • Sales: Find similar deals
  • FAQ: Match questions to answers

Search Configuration​

Similarity Threshold​

ValueMeaning
0.9+Very similar (exact match)
0.7-0.9Relevant
0.5-0.7Somewhat related
Below 0.5Probably not relevant

Filtering​

Filter results by metadata:

  • Category
  • Date range
  • Custom tags

Vector Sync​

CRM data syncs to knowledge bases via:

  • Database triggers
  • Background processing
  • Near real-time updates

Analytics​

Track KB performance:

  • Searches - Total queries
  • Hit rate - Found relevant results
  • Top queries - Most common searches
  • Missing queries - No results found

Best Practices​

  1. Organize by topic - Multiple focused KBs
  2. Keep content fresh - Update regularly
  3. Use metadata - Enables filtering
  4. Monitor searches - Find content gaps
  5. Test thoroughly - Verify search quality

Related & Next Steps​

Last updated on