SharePoint Embedded copilot Advanced Topics Overview

This advanced guide covers how the semantic index powers Retrieval-Augmented Generation (RAG) to provide accurate, context-aware AI responses. We’ll explore how these concepts work together to ensure your copilot retrieves relevant information from your data and returns grounded answers.

Caveats

Configuration

Required ContainerType Configuration

DiscoverabilityDisabled

This flag prevents copilot from discovering drive items in the specified container type. If you have an existing container type and are setting this value to false, please wait 24 hours to ensure the container type configuration is fully propagated before creating a new container, uploading files there, and trying out copilot on folders/files of that new container.

The following is an example of how to set the flag to false with Set-SPOContainerTypeConfiguration

Set-SPOContainerTypeConfiguration -ContainerTypeId 4f0af585-8dcc-0000-223d-661eb2c604e4 -DiscoverabilityDisabled $false

Discoverability can also be disabled using the Visual Studio Code SharePoint Embedded extension

CSP Policies

The Content-Security-Policy (CSP) for embedded chat hosts ensures that only specified hosts can load the chat component. This helps in securing the application by restricting which domains can embed the chat component.

It is intended to allow consuming tenant SPE admins to set an allowlist of hosts that they will allow to embed the SPE copilot in an iFrame. Specifically, the value they set here will be used in a Content-Security-Policy header as a frame-ancestors value.

Below are example commands to use the Connect to SharePoint Online using PowerShell commands:

• Set-SPOApplication to set the CopilotEmbeddedChatHosts property.
• Get-SPOApplication to get the CopilotEmbeddedChatHosts property.

# Note this MUST be run in Windows PowerShell. It will not work in PowerShell.
Import-Module -Name "Microsoft.Online.SharePoint.PowerShell"
Connect-SPOService "https://<domain>-admin.sharepoint.com"
# Login with your admin account.
...

Set-SPOApplication -OwningApplicationId 423poi45 -CopilotEmbeddedChatHosts "http://localhost:3000 https://contoso.sharepoint.com https://fabrikam.com" 

# This will set the container type configuration “CopilotEmbeddedChatHosts” accordingly. 
...

Get-SPOApplication -OwningApplicationId <OwningApplicationId> | Select-Object CopilotEmbeddedChatHosts

OwningApplicationId             : <OwningApplicationId>
OwningApplicationName           : SharePoint Embedded App
Applications                    : {<OwningApplicationId>}
SharingCapability               : ExternalUserAndGuestSharing
OverrideTenantSharingCapability : False
CopilotEmbeddedChatHosts        : {http://localhost:*}

Optional Configuration

Authentication and 3P Cookies

The iFrame used by SharePoint Embedded copilot attempts to authenticate using third-party cookies. If third-party cookies are disabled in the user's browser, the iFrame will not be able to authenticate automatically. In such cases, a popup will be displayed prompting the end user to log in manually. This ensures that the authentication process can still be completed even when third-party cookies are not available.

Advanced Topics

Application Scoping

Application scoping in SharePoint Embedded copilot (SPE copilot) involves defining the boundaries and context within which the tool operates, ensuring its features and capabilities are tailored to meet the specific needs of different applications. This process helps customize the copilot's functionality, making it more effective and relevant for various use cases.

When SPE copilot users query the LLM, it will only have access to files that the User+Application have access to. The effective permissions for the copilot session will be the intersection of your SharePoint Embedded application's permissions and the user's permissions.

Information Architecture

Files in SharePoint Embedded are naturally semantic indexed. This semantic index underpins retrieval augmented generation (RAG) workflows by providing relevant context from your stored content at query time. In essence, it grounds the AI responses, ensuring they directly reference accurate information in your containers rather than relying on general knowledge alone.

With SharePoint Embedded copilot, you can further ground the large language models (LLM) reponses on specific files or drive items..

Semantic index

Learn more about semantic index for Microsoft 365 Copilot here

The semantic index allows for quick and accurate searches based on data similarity. This means it can find the most relevant information not just by exact matches, but also by understanding the context and meaning.

Retrieval-Augmented Generation (RAG)

RAG relies on having relevant source materials stored in a repository, which can be queried at runtime​, data is retrieved from the index and is used to augment the prompt sent to the large language model (LLM)​:

• Treat data sources as knowledge without having to train your model​
• Uses search (retrieval) results as additional context in your prompt​
• Generates the output using the prompt and the supplied context

The LLM uses the data to inform and construct the response.

​

Grounding

Grounding in the context of SPE copilot refers to the process of providing input sources to the large language model (LLM) related to the user's prompt. This helps improve the specificity of the prompt and ensures that the responses are relevant and actionable to the user's specific task. The data the copilot is grounded on will be on the contents of the container type in the copilot application. Behind the scenes SPE copilot uses M365 Copilot, learn more about it's architecture here

Scoping your copilot to specific content

SharePoint Embedded (SPE) copilot has the ability to restrict the data sources it has access to, below are provided types, and this example shows how to configure the SDK

export type IDataSourcesProps =
  | IFileDataSource
  | IFolderDataSource
  | IDocumentLibraryDataSource
  | ISiteDataSource
  | IWorkingSetDataSource
  | IMeetingDataSource;

export enum DataSourceType {
  File = 'File',
  Folder = 'Folder',
  DocumentLibrary = 'DocumentLibrary',
  Site = 'Site',
  WorkingSet = 'WorkingSet',
  Meeting = 'Meeting'
}

Supported document types for scoping

Reference - File Formats Support By copilot

Documents: PDF, DOCX, XLSX, PPTX

Text-based Files: RTF, TXT, CSV, LOG, INI, CONFIG

Audio: WAV

Programming Languages: PY, JS, JSX, JAVA, PHP, CS, C, CPP, CXX, H, HPP, M, COFFEE, DART, LUA, PL, PM, RB, RS, SWIFT, GO, KT, KTS, R, SCALA, T, TS, TSX

Shell Scripts: BASH, SH, ZSH

Markup and Documentation: HTML, CSS, MD, RMD, TEX, LATEX

Database Languages: SQL

Data Serialization Formats: IPYNB, JSON, TOML, YAML, YML

Language/Locale

The copilot iframe dynamically loads localization settings to ensure that the chat interface is displayed in the appropriate language. These settings are derived from SharePoint, which provides a comprehensive set of localization options.

When the copilot iframe is initialized, it retrieves the current localization settings from SharePoint. These settings dictate the language and regional preferences for the chat interface, ensuring that all UI elements, messages, and interactions are presented in the user's preferred language. This seamless integration with SharePoint's localization framework allows copilot to provide a consistent an

You can have this localized by setting your language options in the SharePoint account settings: Change your personal language and region settings - Microsoft Support note, if your M365 setting is different from your Sharepoint account langauge settings it will take precedence, you can change your M365 language settings here: Change your display language in Microsoft 365

An additional locale option can be passed in through the ChatLaunchConfig to further set the language the copilot will respond in:

 const [chatConfig] = React.useState<ChatLaunchConfig>({
        header: ChatController.instance.header,
        theme: ChatController.instance.theme,
        zeroQueryPrompts: ChatController.instance.zeroQueryPrompts,
        suggestedPrompts: ChatController.instance.suggestedPrompts,
        instruction: ChatController.instance.pirateMetaPrompt,
        locale: "en",
    });

Locale Options

Here are some examples of locale options you can use: