In this article we talk about how the knowledgebase section in FlowGent works.

Understanding How Our Knowledgebase Works

When we talk about "training" an AI agent on your data, we're not actually rebuilding or retraining the entire AI model. Instead, we're creating what's called a Retrieval-Augmented Generation (RAG) system that allows the AI to access and reference your specific information when answering questions.

Think of it like giving a human assistant access to your company's documents and files. The assistant already knows how to read and understand language—they just need access to your specific information to provide relevant answers.

How It Works (In Simple Terms)

1. We take your information (from websites, uploaded files, or direct text)
2. The system breaks this information into manageable pieces
3. These pieces are stored in a searchable format
4. When a user asks a question, the system retrieves the most relevant pieces
5. The AI uses these pieces to formulate an accurate, informed response

This approach means your AI agent can draw from your specific knowledge without needing to modify its core understanding of language.

The Technical Details (For Those Interested)

Behind the scenes, here's what happens:

• Chunking: We break your content into smaller, digestible pieces (or "chunks").
• Vector Embeddings: Using OpenAI's embedding technology, we convert these chunks into numerical representations (vectors) that capture their meaning.
• Similarity Search: When a question is asked, we find the chunks whose vectors are most similar to the question's vector.
• Contextual Response: The AI then uses these relevant chunks as context to generate an accurate response.

Benefits of This Approach

• Cost-efficiency: We only retrieve and process the relevant information, keeping costs lower
• Performance: The AI isn't overwhelmed with irrelevant information
• Flexibility: Your knowledge can be updated without needing to retrain the entire model
• Accuracy: Responses are grounded in your specific information

Instructing Your Agent When to Use the Knowledgebase

You can customize when your AI agent references the knowledgebase:

• On specific topics: "Reference the knowledgebase when questions about our products are asked"
• For in-depth queries: "Use the knowledgebase when users ask detailed technical questions"
• Always: "Always check the knowledgebase before responding to any question"
• User-triggered: "Reference the knowledgebase only when the user specifically requests it"

Data Sources For Your Knowledgebase

Website Scraping

We can extract information directly from your website:

• Free Trial: Up to 15 pages
• Paid Plan: Up to 500 pages
• Crawl Depth: Choose from 1-6 (minimum 1, maximum 6)
  • Crawl depth refers to how many links deep we follow from your starting URL. A depth of 1 only captures the specific URL you entered. A depth of 2 also captures the pages directly linked from your starting page, while a depth of 6 follows links through five levels of pages.

Exclude Paths: You can specify paths to exclude, such as /blog/ or /docs/old/. Simply enter these paths separated by commas or press Enter after each one.

File Upload

Upload your existing documentation:

• Supports PDF documents
• Text files

Plain Text

Enter information directly as plain text for immediate inclusion in the knowledgebase.

System Instructions vs. Knowledgebase: When to Use Which

Use System Instructions When:

• You need to define the agent's persona or behavior
• You're providing rules about how the agent should interact
• You're setting guardrails or constraints
• You have a small amount of critical information (less than a page)
• You need the information to be always considered, regardless of query

Use the Knowledgebase When:

• You have large volumes of information (documentation, FAQs, product details)
• The information changes frequently
• You want selective retrieval based on relevance
• You're working with structured data like pricing, specifications, or procedures
• You want to reduce costs by only retrieving what's needed

Coming Soon: Additional Integrations

We're actively working on expanding our integration options to include:

• Google Drive: Connect your documents and spreadsheets
• Notion: Integrate your Notion workspace

Would you like to see these integrations sooner? Let our AI agent know in the app or on our website that you're interested in these features. Your feedback helps us prioritize our development roadmap!