Knowledge Bases

Knowledge Bases in OneByZero Neo provide your AI agents with access to organizational information, documents, and data that enable them to deliver accurate, contextually relevant responses. Neo supports multiple knowledge bases per workspace, and these knowledge bases can be reused across multiple agents, making them a powerful shared resource for your AI ecosystem.

Knowledge Bases Overview

Knowledge Base Types

OneByZero Neo supports two types of knowledge bases:

  • In-House Knowledge Bases - Powered by OpenSearch, these knowledge bases are fully managed within the Neo platform and provide a straightforward way to store and retrieve document content.

  • Bedrock Knowledge Bases - Integration with AWS Bedrock Knowledge Bases, offering enterprise-grade vector storage and retrieval capabilities backed by Amazon’s infrastructure.

You can switch between viewing In-House and Bedrock knowledge bases using the tabs at the top of the Knowledge Bases page.

Reusability Across Agents

One of the key advantages of knowledge bases in Neo is their reusability. A single knowledge base can be connected to multiple agents within a workspace, providing several benefits:

  • Consistency - All agents accessing the same knowledge base work from the same information, ensuring consistent responses across different touchpoints

  • Efficiency - Maintain documents in one place rather than duplicating content across agents

  • Simplified Updates - When information changes, update the knowledge base once and all connected agents automatically access the updated content

  • Cost Optimization - Avoid redundant storage and ingestion costs by sharing knowledge bases

For example, a knowledge base containing company policies can be used by both a customer support agent and an internal HR assistant, ensuring both provide consistent policy information.

In-House Knowledge Bases

In-House knowledge bases are backed by OpenSearch and provide a fully integrated solution for document storage and retrieval within Neo.

Managing Documents

Clicking on an In-House knowledge base card opens the details view where you can manage the documents contained in the knowledge base.

In-House Knowledge Base Details

The details view displays:

  • Description - The knowledge base description explaining its purpose and contents

  • Type - Indicates IN-HOUSE for OpenSearch-backed knowledge bases

  • Ingestion Status - Shows whether documents have been successfully ingested (e.g., “ingestion success”)

  • Documents Table - Lists all documents with their name, upload date, and availability status

Action Buttons

  • Add Document - Upload new documents to the knowledge base

  • Ingest - Trigger the ingestion process to index newly uploaded documents into the vector store

Uploading Documents

Click Add Document to upload files to your knowledge base.

Upload Files Dialog

The upload dialog supports:

  • Supported file formats - .pdf, .docx, .pptx, .xlsx, .md, .json

  • Upload limits - Maximum 10 files per upload, 10 MB per file

After uploading, click Ingest to process the documents and make them available for retrieval by agents.

Bedrock Knowledge Bases

Bedrock knowledge bases integrate with AWS Bedrock’s managed knowledge base service, providing enterprise-grade vector storage and retrieval.

Bedrock Knowledge Base Configuration

Creating a Bedrock Knowledge Base

When creating a Bedrock knowledge base, you configure:

Basic Information

  • Name - A unique identifier for the knowledge base

  • Description - A description of the knowledge base’s purpose and contents

Embedding Configuration

  • Embedding Model Id - Select the embedding model used to convert documents into vector representations

Storage Configuration

  • Storage Type - Choose the vector store backend (e.g., OpenSearch Managed Cluster)

Data Source

Configure where the knowledge base retrieves its source documents from.

Metadata Files for Bedrock

Bedrock knowledge bases support enhanced metadata through companion metadata files. For any document you upload, you can include a metadata file following the naming pattern:

<filename>.metadata.json

For example, if you upload policy_document.pdf, you can include policy_document.pdf.metadata.json to attach metadata to that document. This metadata is then associated with each chunk in the vector store, enabling:

  • Filtered retrieval based on metadata attributes

  • Enhanced search relevance

  • Document categorization and tagging

  • Custom attributes for your specific use case

Updating Knowledge Bases

Knowledge bases can be updated through the UI by:

  1. Adding new documents - Upload additional files to expand the knowledge base

  2. Removing documents - Delete outdated or incorrect documents

  3. Re-ingesting - Trigger re-ingestion after making changes to ensure the vector store is updated

For In-House knowledge bases, the workflow is:

  1. Navigate to the knowledge base details

  2. Click Add Document to upload new files

  3. Click Ingest to process and index the new content

  4. Verify the ingestion status shows success

Best Practices

Organize by Domain

Create separate knowledge bases for different domains or topics. This improves retrieval accuracy and makes management easier.

Keep Content Current

Regularly review and update knowledge base content to ensure agents provide accurate information.

Use Descriptive Names

Name knowledge bases clearly to indicate their purpose and contents, making it easier for team members to select the right knowledge base for their agents.

Monitor Ingestion Status

After uploading documents, verify that ingestion completes successfully before relying on the content in production agents.

Leverage Metadata

For Bedrock knowledge bases, use metadata files to enrich your documents with additional context that can improve retrieval quality.