Skip to main content
Rather than hardcoding reference material into an agent’s instructions, you attach a datasource so the agent can look up current information at query time — from uploaded files, crawled web pages, pasted text, or a connected Google Drive folder.

Core Concept

DATASOURCE = Knowledge Base + Search + Retrieval
  • Datasource contents:
    • Files (PDFs, Docs)
    • Web Content (URLs)
    • Manual Text
    • Google Drive Files
    • Metadata & Indexing
  • Vector Search — searches across the datasource contents
  • Relevant Context — returned to the agent

Datasource Capabilities

Every datasource can:

Store Knowledge

Hold large amounts of structured and unstructured information

Enable Search

Find relevant information using semantic search

Scale Content

Support multiple files, URLs, and manual content entries

Connect to Agents

Be referenced by multiple agents for shared knowledge

Key Features

Knowledge Storage

Datasources can store various types of content:
  • Files: PDF documents, Word documents, text files, spreadsheets
  • Web Content: URLs, entire websites, sitemaps
  • Manual Content: Text entered directly in the interface
  • Cloud Storage: Google Drive files and folders
  • Metadata: Custom tags, descriptions, document properties

Vector Search & Retrieval

Shared Access

Datasources can be:
  • Assigned to multiple agents - Share knowledge across your team
  • Updated centrally - Changes reflect immediately for all agents
  • Managed by admins - Control access and permissions
  • Organized by project - Keep knowledge organized by context

Datasource Types

Manual Datasources

Self-managed knowledge bases where you add content directly:
  • Upload Files
  • Add URLs
  • Enter Text Content
Best for:
  • Company policies and documentation
  • Custom training materials
  • Internal knowledge bases
  • Proprietary information

Google Drive Datasources

Connected directly to Google Drive folders:
  • Authenticate with Google
  • Select Folders
  • Auto-sync Files
Best for:
  • Shared team documents
  • Living documentation
  • Multi-author content
  • Automatic updates

Common Use Cases

Customer Support

Store FAQ, knowledge base, and troubleshooting guides

Product Knowledge

Maintain documentation, API specs, and feature guides

Company Policies

Reference HR policies, procedures, and guidelines

Training Materials

Store courses, onboarding docs, and best practices

How Agents Use Datasources

Configuration

When creating an agent:
  1. Select Datasources - Choose which knowledge bases the agent can access
  2. Configure Search - Set retrieval parameters (Top K, reranking, etc.)
  3. Test Queries - Verify the agent can find relevant information

Runtime Behavior

When an agent receives a question:
  1. Question — e.g. “How do I reset my password?”
  2. Agent searches datasources
  3. Finds — e.g. “Password Reset Guide.pdf” and FAQ entry
  4. Retrieves top 10 relevant sections
  5. Includes in context for response
  6. Agent responds with confidence

Datasource Organization

Metadata

Each datasource can have:
  • Name: Descriptive identifier
  • Description: Purpose and contents
  • Tags: Categorization (e.g., “public”, “beta”, “deprecated”)
  • Visibility: Public or private
  • Ownership: Which team member created it

Resources

Each datasource contains multiple resources:
  • Files: Uploaded documents
  • URLs: Web pages and sitemaps
  • Content: Manual text entries
  • Metadata: File properties, indexing status

Shared Datasources

Team Collaboration: Create shared datasources that multiple agents can reference. This ensures consistent information and reduces duplication.
  • Centrally managed by admins
  • Accessible to multiple agents
  • Updated once, used everywhere
  • Audit trail of changes

Best Practices

Organized Structure

Group related content into separate datasources

Clear Naming

Use descriptive names for datasources and files

Regular Updates

Keep information current and remove outdated content

Test Retrieval

Verify agents can find the information they need

Content Quality

  • Clear Structure: Use headings, lists, and sections
  • Complete Information: Provide full context, not snippets
  • Consistent Formatting: Standardize document structure
  • Remove Duplicates: Avoid storing the same content multiple times

Datasources vs Agents


Next Steps