Grounded Knowledge Indexes
Grounded Knowledge Indexes enable your AI agents to access and search through your organization's documents and knowledge bases. This allows agents to provide contextually relevant answers grounded in your actual data.
What is Grounded Knowledge?
Grounded Knowledge uses a RAG (Retrieval-Augmented Generation) architecture that combines:
- Retrieval - Finding relevant documents from a knowledge base using semantic search
- Augmentation - Adding the retrieved context to the AI prompt
- Generation - Using an LLM to generate answers based on the context
This approach allows AI agents to:
- Answer questions using your organization's specific documents
- Provide accurate, up-to-date information from your knowledge base
- Cite sources for their responses
- Avoid hallucinations by grounding responses in real data
Components
A Grounded Knowledge Index consists of four Azure building blocks:
| Component | Purpose |
|---|---|
| Search Index | Stores document embeddings with semantic search configuration |
| Knowledge Source | Links the search index to agentic retrieval with searchable field definitions |
| Knowledge Base | Aggregates one or more Knowledge Sources; Azure handles multi-source fan-out internally |
| Chat Completion Model | Azure OpenAI deployment (e.g., gpt-4.1-mini) that synthesizes contextual responses |
Getting Started
To set up a Grounded Knowledge Index:
- Navigate to Build > Knowledge in the Control Bridge admin console
- Click Create New Index
- Follow the Combined Setup Wizard to select your index types and configure Azure resources
The Combined Setup Wizard allows you to provision multiple index types in a single session. For example, you can create both an Email Archive and a Corporate Knowledge index at the same time, sharing the same Azure providers.
Assigning Indexes to Agents
Once a Grounded Knowledge Index is provisioned and active, it automatically appears as a tool that can be assigned to agents. Agents with access to an index can:
- Search through indexed documents
- Retrieve relevant passages for context
- Generate answers based on the retrieved information
See Tools for more information on assigning tools to agents.
Index Types
Grounded Knowledge Indexes support the following types of knowledge bases:
| Type | Purpose |
|---|---|
| Email Archive | Index and search historical email archives. Used as the Email History Index in Account Settings. |
| Corporate Knowledge | Index internal corporate documentation including employee handbooks, policies, procedures, contracts, and benefits information. Used as the Corporate Knowledge Index in Account Settings. |
| SharePoint Documents | Index documents from SharePoint Online libraries with automatic sync. |
| Google Drive Documents | Index documents from Google Workspace Shared Drives with OAuth integration. |
| Activity Log Search | Index agent execution activity summaries for operational insight. |
Email Archive
Email Archive indexes are designed for searching historical email communications. When created, they are automatically configured as the tenant's Email History Index, enabling the Index Knowledge tool to mine and index emails.
Key Features:
- AI-powered email mining automatically identifies valuable emails from shared mailboxes
- Value assessment scores each email's knowledge potential (0-1 scale)
- Topic classification categorizes emails into your defined taxonomy
- Approval workflow ensures quality control before permanent indexing
- Incremental sync efficiently processes only new emails via Microsoft Graph delta queries
- Scheduled scanning supports daily automated scans for mailbox knowledge sources
Best for: Customer service teams, executive assistants, and anyone who needs to search past communications for context.
See Email Indexing for Search for detailed setup and configuration instructions.
Corporate Knowledge
Corporate Knowledge indexes provide a centralized, searchable repository for internal company documentation. This enables AI agents to answer questions about company policies, procedures, and guidelines with accurate, citation-backed responses.
Best for:
- HR Teams - Answer questions about employee handbooks, benefits, PTO policies
- Finance Teams - Reference payroll processes, expense policies, corporate contracts
- Operations Staff - Access vendor agreements, SLAs, emergency procedures
- Specialist Agents (SMEs) - Query corporate knowledge to provide grounded responses
See Corporate Knowledge Setup for detailed setup instructions.
SharePoint Documents
SharePoint Document indexes connect to SharePoint Online libraries and automatically sync documents for AI-powered search. Documents are indexed with AI-generated summaries and metadata extraction.
Best for: Organizations with corporate policies, procedures, and documentation stored in SharePoint.
See SharePoint Knowledge Sources for detailed setup instructions.
Google Drive Documents
Google Drive indexes connect to Google Workspace Shared Drives and sync documents with OAuth integration. Supports a variety of document types including Google Docs, PDFs, and Office files.
Best for: Organizations using Google Workspace for document management and collaboration.
See Google Drive Knowledge Sources for detailed setup instructions.
Activity Log Search
Activity Log indexes capture agent execution summaries, enabling you to search through historical agent activity for operational insights and audit purposes.
Best for: Operations teams that need visibility into what agents have done, and agents that need to reference past activity for context.
Multi-Source Knowledge Bases
A Knowledge Base can aggregate multiple Knowledge Sources, each backed by its own Search Index. This allows you to combine different types of content (e.g., emails and SharePoint documents) into a single searchable endpoint. Azure handles multi-source fan-out internally during retrieval.
You can manage the sources within a Knowledge Base from the index details page. See Knowledge Bases for more details.
Scheduled Scanning
For email-based knowledge sources, you can configure scheduled scanning to keep your indexes up to date automatically:
- Daily scanning runs automatically every 24 hours
- Incremental (delta) scanning processes only new or changed emails
- Manual scanning can be triggered at any time from the knowledge source detail page
Monitor active and completed scans from the Scan Activity page.
Managing a Grounded Knowledge Index
After creating an index, you can view and manage it from the index details page. The details page has three tabs:
Overview Tab
The Overview tab shows:
- Stats Summary Card - At-a-glance health and usage metrics for the index (see below)
- Configuration Summary - Name, type, and description
- Azure Providers - Connected OpenAI and Search service details
- Component Status - Status of the index, knowledge source, and knowledge base
Stats Summary Card
Near the top of the Overview tab, a four-tile stats card gives operators an immediate read on whether the index is active and how much data it covers:
| Tile | Description |
|---|---|
| Total Queries | Cumulative number of times the index has been queried by agents |
| Last Queried | Relative time since the last query (e.g., "3 hours ago"), with the exact timestamp on hover. Shows "Never" if the index has not yet been queried. |
| Total Documents | Sum of all documents across every knowledge source in the index |
| Total Storage | Combined storage used by all sources, displayed in human-readable units (KB, MB, GB) |
A brand-new index with no traffic shows 0 queries, Never for last queried, and 0 B storage - all normal before content is indexed and agents start searching.
Query Activity Card
Below the stats summary, a Query Activity (30 days) card gives operators a rolling view of how the index is being used and how healthy it is:
| Section | Description |
|---|---|
| Sparkline | Area chart showing queries per day for the last 30 days. Hover over any point to see the date and count. |
| p95 Latency | 95th-percentile retrieval time in milliseconds over the window - the slowest "normal" query. |
| Avg Latency | Mean retrieval time in milliseconds. Compare with p95 to spot outliers. |
| Success Rate | Percentage of queries that completed without an error. |
| Errors | Total number of failed queries in the window. Shown in red when non-zero. |
| Most Recent Error | If any errors occurred, the most recent error message (truncated to 200 characters) and its relative timestamp. |
| Top Users | Up to three users with the most queries in the window, with their query counts. |
If the index has had no queries in the last 30 days, the card shows "No queries in the last 30 days" instead of charts and stats - this is normal for a newly provisioned or inactive index.
Topic Taxonomy Tab
For Email Archive and Corporate Knowledge type indexes, you can configure a topic taxonomy to categorize indexed content:
- Create from Template - Import a predefined topic structure (recommended for email archives)
- Add Topics Manually - Create custom topic categories
- Hierarchical Structure - Topics can have parent/child relationships
- Search & Filter - Find specific topics in large taxonomies
Topic taxonomies help agents classify content and enable filtering by topic when searching.
Test Query Tab
Test your index's search functionality:
- Enter a search query
- Click Test Query
- View the results including:
- Synthesized Answer - The AI-generated response based on retrieved documents
- Token Usage - Prompt tokens, completion tokens, and total tokens used
- References - Footnote-style list of source documents with reference IDs, document titles, and reranker scores
Account Settings Integration
Email History Index
Email Archive indexes are automatically configured as the Email History Index for email mining when provisioned. This means:
- Create an Email Archive Grounded Knowledge Index using the Setup Wizard
- The index is automatically set as the Email History Index in Account Settings
- Knowledge source scans will work immediately without additional configuration
If you need to change which index is used for email history, you can manually select a different one in Manage > Account > Settings.
Corporate Knowledge Index
Similarly, Corporate Knowledge indexes are automatically designated as the tenant's Corporate Knowledge Index in Account Settings when provisioned:
- Create a Corporate Knowledge Grounded Knowledge Index using the Setup Wizard
- The index is automatically set as the Corporate Knowledge Index in Account Settings
- Specialist Agents (SMEs) with appropriate access can query the corporate knowledge base immediately