Knowledge Base

Create a knowledge base to give your agents access to grounded information.

Knowledge Base enables your agents to retrieve accurate, sourced information from your documents. This is essential for RAG (Retrieval-Augmented Generation) workflows.

Note:

Knowledge Base setup is optional. Add it when you need document-grounded responses.

When to Use Knowledge Base

Use CaseExample
Q&A SystemsAnswer questions from company documentation
Research AssistantsSearch across uploaded materials
Grounded ResponsesEnsure factual accuracy with citations
Domain ExpertiseTrain agents on specialized knowledge
  1. Open the sidebar menu
  2. Click Knowledge Base to expand
  3. Select Knowledge Bases

You'll see your existing knowledge bases (if any) and options to create new ones.

Knowledge Base Nav Fig : Knowledge Base landing window

Create a Knowledge Base

The creation process has 3 steps:

  1. Create Empty Knowledge Base — Configure name, provider, and settings
  2. Document Upload — Add your files
  3. Execute & Finish — Process and index documents

Step 1: Configure Knowledge Base

FieldDescription
NameDescriptive name (e.g., "Quarterly Earnings Report")
Knowledge Base ProviderSelect Amazon Bedrock or Milvus
Embedding ModelModel for vector embeddings (e.g., Titan G1 - Text v2.0)
Data SourceWhere raw files are stored
MetadataOptional key-value pairs for filtering

Knowledge Base Providers

ProviderStatusStorageVector DBBest For
Amazon BedrockProductionAmazon S3Bedrock KBEnterprise, AWS users
MilvusIn DevelopmentNFS StorageMilvusSelf-hosted, custom setups

Amazon Bedrock Setup

  1. Select Amazon Bedrock as the provider
  2. Choose embedding model: Titan G1 - Text v2.0 (recommended)
  3. Data Source: Amazon S3 (automatically configured)
  4. Add optional metadata fields for document filtering
  5. Click Create and Continue

Knowledge Create Fig : Knowledge Base create

Note:

Amazon Bedrock Knowledge Base uses S3 to store raw files and Bedrock's managed vector store for embeddings.

Milvus Setup (Coming Soon)

  1. Select Milvus as the provider
  2. Configure NFS storage path for raw files
  3. Milvus connection settings for vector storage

Step 2: Add Files

After creating the knowledge base, you'll see the management view with tabs: Overview, Files, Analytics, Settings.

Knowledge Create files add Fig : Knowledge Base with initial files

Knowledge Finish Setup Fig : Knowledge Base Finish Setup

Supported File Types

TypeExtensions
DocumentsPDF, DOCX, TXT, MD
DataJSON, CSV, YAML

Upload Process

  1. Click Add Files button
  2. Drag files into the upload area or click to browse
  3. Files are staged (uploaded but not yet indexed)

Step 3: Sync Knowledge Base

After adding files, you must sync to create vector embeddings:

  1. Click Sync Knowledge Base button
  2. Wait for processing to complete
  3. Use Check Knowledge Base Status to monitor progress

Knowledge Finish Sync Fig : Knowledge Base Sync for embeddings creation

Note:

Uploaded files are staged first. They won't be searchable until you sync the knowledge base. After sync it will take few minutes to create embeddings.

Document Status

StatusMeaning
StagedFile uploaded, not yet indexed
ProcessingCreating vector embeddings
ProcessedFile indexed and searchable
FailedProcessing error (check format)

Management Actions

ActionDescription
Add FilesUpload new documents
Sync Knowledge BaseProcess staged files into vectors
Check Knowledge Base StatusView sync progress
DeleteRemove the knowledge base

Note:

You can add multiple files before syncing. Sync processes all staged files at once.

Test in Playground

Before connecting to workflows, test your knowledge base:

  1. Go to Knowledge Base → Playground
  2. Select your knowledge base
  3. Enter a test query
  4. Review retrieved results and relevance

This helps verify your documents are properly indexed.

Knowledge Playground Fig : Knowledge Base Playground

Note:

You must have configured the default model provider from Configurable Component

Connect to Workflow

Add as Agent Tool

  1. Open your workflow in the Flow Editor
  2. Select an Agent node
  3. In the agent's Tools section, add Knowledge Base
  4. Select your knowledge base from the dropdown
  5. Configure retrieval settings:
SettingDescription
Top KNumber of results to retrieve
Similarity ThresholdMinimum relevance score

Knowledge Base with Workflow Fig : Knowledge Base inside workflow

How It Works

User Query → Agent → Knowledge Base Tool → Retrieve Documents → Generate Response

The agent automatically:

  1. Converts the query to a vector embedding
  2. Searches for similar document chunks
  3. Uses retrieved context to generate accurate responses

Best Practices

  • Chunk size: Smaller chunks (500-1000 tokens) for precise retrieval
  • Document quality: Clean, well-structured documents index better
  • Naming: Use clear names to identify knowledge bases easily
  • Updates: Re-upload documents when content changes
  • Testing: Always test queries in Playground first

Troubleshooting

IssueSolution
"No results found"Check if documents are in "Ready" status
"Irrelevant results"Adjust similarity threshold or re-index
"Processing stuck"Check file format and size limits

Next: MCP Servers (Optional)