Magnate

Knowledge Base

The Knowledge Base is the brain of your customer-facing AI agent. It is where you provide the context, business rules, and information that the AI uses to accurately answer customer inquiries in the Inbox using Retrieval-Augmented Generation (RAG).

The Knowledge Base interface is divided into three core tabs: Sources, Gaps, and Test AI.

1. Adding Knowledge Sources

The Sources tab is where you upload and manage the data the AI learns from. Click the Add Source button to ingest new knowledge.

You can train the AI using 5 distinct formats:

  • Website: Provide a URL, and the system will crawl and extract text from the webpage.
  • File: Upload documents (PDF, DOCX) containing company policies, product manuals, or catalogs.
  • Q&A: Input specific Frequently Asked Question pairs.
  • Text: Directly paste or type raw text snippets.
  • AI Correction: Provide specific overrides or corrections to guide the AI's behavior on nuanced topics.

2. Managing & Indexing

Once you add a source, the system breaks it down into semantic "chunks" for the AI to understand. You can track this process via the status indicators in the Sources tab:

  • Pending: The source is queued for processing.
  • Extracting: The system is currently reading and vectorizing the data.
  • Indexed: The source is fully processed and the AI is actively using it to answer questions.
  • Failed: The system could not read the source (e.g., an inaccessible URL or corrupted file).

Important: The AI can only answer questions using sources that have successfully reached the Indexed status.

3. Closing Knowledge Gaps

The Gaps tab is a powerful continuous learning feature.

When the AI encounters a customer question in the Inbox that it cannot answer confidently using its existing knowledge, it logs an "Unresolved Gap."

  • Identify blind spots: View a prioritized list of unresolved queries, including how many times customers have asked them (Occurrence Count) and when they were last seen.
  • Resolve the gap: You can resolve a gap by uploading a new source or typing a direct answer. Once resolved, the AI will know exactly how to handle that question the next time it arises.

4. Testing Your AI

The Test AI tab is your testing playground. Before putting your AI agent in front of real customers, you can simulate questions here.

  • Search Queries: Type a sample question into the search bar.
  • Semantic Results: The system will return the exact "chunks" of text the AI retrieved from your knowledge base to formulate its answer.
  • Confidence Scores: View the relevance score of each retrieved chunk to ensure the AI is pulling the most accurate and contextually appropriate information.

By regularly monitoring Gaps and Testing your AI agent, you ensure your automated agent remains highly accurate and continually improves!