Perplexity's SharePoint connector and OneDrive connector enable Enterprise Pro and Enterprise Max users to search their organization's SharePoint sites and OneDrive files directly through AI-powered natural language queries.
We have recently introduced a hybrid search architecture that addresses both data security concerns and scalability limitations through two distinct search modes: High-Precision Search (indexing-based) and Standard Search (API-based).
Search Architecture: Two-Tier Approach
Standard Search (API-Based)
How It Works:
-
Queries SharePoint/OneDrive directly via Microsoft's Search API at query time
-
No file copies stored in Perplexity infrastructure, aside from results found from queries (can be mitigated by data retention policy settings)
-
Searches across the user's entire SharePoint/OneDrive drive without file count limitations
-
Available to all Enterprise Pro/Max users immediately upon connector activation
Security Model:
-
Zero Data Retention: Files are not copied or stored in Perplexity systems, aside from results found from queries (can be mitigated by data retention policy settings)
-
Real-time Permissions: Respects SharePoint's/OneDrive’s native access controls dynamically
-
Minimal Data Copying: Only citation snippets included in answers are retained
-
No Model Training: Synced files are never used to train AI models
-
Immediate Access Revocation: When SharePoint/OneDrive permissions change, access is immediately reflected in Perplexity
Use Case: Organizations requiring maximum data privacy and minimal data footprint, especially for searching across millions of files at enterprise scale.
High-Precision Search (Indexing-Based)
How It Works:
-
Users select specific files/folders to sync for local indexing in Perplexity
-
Files are downloaded, parsed, and stored in dedicated AWS S3 buckets with vector embeddings in Vespa
-
Enables deeper semantic analysis and more comprehensive answers
-
File Limits: 500 files per project (Enterprise Pro), 5,000 files per Project (Enterprise Max)
-
Total User Limits: 15,000 files (Enterprise Pro), 50,000 files (Enterprise Max)
Security Model:
-
Dedicated Storage: Each organization's files stored in isolated AWS S3 "folders" with unique namespaces in Vespa vector storage
-
Encryption: AES-256 encryption at rest, TLS encryption in transit
-
Role-Based Access Control (RBAC): Minimum privilege access enforced across all systems
-
No Model Training: Synced files are never used to train AI models
-
Automatic Sync: File changes/deletions in SharePoint/OneDrive are automatically reflected in Perplexity
Use Case: Teams requiring maximum answer accuracy for frequently accessed documents, project-specific file collections, or collaborative projects with curated content.
Backend Architecture & Data Flow
Connection & Authentication
-
Admin Enablement: Organization admins enable SharePoint/OneDrive connector in Permissions settings
-
User Authentication: Users authenticate via OAuth 2.0 through Microsoft Entra (Azure AD)
-
Site Selection: Users select specific SharePoint sites to connect
-
Admin Consent: Microsoft admins may need to grant organization-wide consent for Perplexity app in Microsoft Entra
When Both Methods Are Active:
-
Perplexity queries both the local index and SharePoint API concurrently
-
Results are re-ranked to prioritize the most relevant sources
-
Citations link directly back to SharePoint for full file access
Indexing Process (High-Precision Search)
File Sync and Storage:
-
File Selection: User selects files/folders through Perplexity UI
-
Download: Files downloaded from SharePoint via Microsoft Graph API
-
Storage: Raw files stored in AWS S3 with dedicated organizational namespaces
-
Parsing: Text extraction from supported formats (PDF, DOCX, XLSX, PPTX, CSV, TXT, MD, JSON)
-
Vectorization: Content converted to embeddings and stored in Vespa vector database
-
Metadata Indexing: File metadata (name, path, permissions) indexed for search retrieval
-
Removal: When a user disconnects SharePoint/OneDrive from Perplexity, they can choose to remove any indexed files
Security & Compliance Infrastructure
Permission Enforcement
SharePoint/OneDrive-Side Permissions:
-
If a user loses access to a file in SharePoint/OneDrive, that file is immediately removed from Perplexity
-
File deletions in SharePoint/OneDrive trigger immediate removal from Perplexity index
-
Users can only search files they have explicit SharePoint/OneDrive permissions to access
Perplexity-Side Permissions:
-
Admins control which users can access connectors via Organization settings
-
Files synced to projects are searchable by Project members, but file content requires SharePoint/OneDrive permissions
-
Session sharing respects organizational sharing policies set by admins
Best Practices
When to Use High-Precision Search (Indexing)
-
Project-specific file collections in projects
-
Frequently accessed knowledge base documents
-
Files requiring deep semantic analysis
-
Collaborative team environments with curated content
When to Use Standard Search (API-Only)
-
Searching across vast file repositories (millions of files)
-
Strict data residency and minimal data copying requirements
-
Exploratory searches across infrequently accessed files
-
Organizations with heightened security/compliance constraints