Internal Connectors

Connecting Your Data Sources to Enterprise Search

Enterprise Search is only as good as the data it can access. Connectors link your business tools—file storage, knowledge bases, and messaging platforms—to Mistral’s semantic index. What you enable, and specifically what you expose to each connector, determines both the relevance of the answers and your acceptable risk level.

Primary Connectors

The Google Workspace connector indexes Google Drive—including documents, spreadsheets, presentations, PDFs, and uploaded files—as well as the text content and formatting of Google Docs. Optionally, depending on your internal policy, it can also index Gmail and its attachments. Its most valuable feature is permission handling: synchronization is bidirectional. If you revoke a collaborator’s access to a Drive folder, that change immediately reflects in Enterprise Search, eliminating the need for duplicate rights management.

The Microsoft connector covers SharePoint Online (sites, document libraries, and lists), OneDrive (personal and shared files), and, optionally, Microsoft Teams (channel messages and files shared in conversations). Configuration relies on the Microsoft Graph API: an administrator grants permissions via Azure AD without accessing the content of individual emails. Coordinate this with your IT team, as it is an Azure administration task, not a simple button click in a console.

The Confluence connector indexes pages, blog posts, comments, attachments, restricted spaces (respecting Confluence permissions), labels, and metadata. Confluence is often the richest source of technical documentation and internal procedures. The connector delivers excellent results on this structured content: a well-maintained “Procedures” space yields actionable answers almost immediately.

The Slack connector indexes messages from public channels, messages from private channels (visible only to authorized members), shared files, and discussion threads. Its value is distinct: it captures informal decisions and expert exchanges that no one formalizes in an official document, such as an explanation shared on a Tuesday night in a technical channel and never documented elsewhere.

Mistral regularly expands its catalog. As of April 2026, Notion (pages and databases), Jira (tickets and projects with descriptions, comments, and attachments), Box (files and folders), and Zendesk (knowledge base articles) are available or in beta. A generic API is also available to connect proprietary sources via a custom connector.

Configure with Discernment

The guiding principle is least privilege: grant only necessary permissions. If you do not need to index emails, do not connect Gmail. The temptation to connect everything “just to see” is strong, but it creates a significant audit burden. Limit indexing to specific Confluence spaces or SharePoint sites rather than the entire instance. Systematically exclude folders containing sensitive personal data, starting with individual HR files and medical records.

Each connector allows you to refine indexing with filtering rules. You can restrict indexing to specific folders or spaces—for example, “Documentation” and “Procedures” while excluding personal spaces. You can exclude file types like video, audio, or executables. You can filter out documents older than a specific date to prevent outdated procedures (such as a 2019 policy) from surfacing. Alternatively, you can index only documents tagged with a “Validated” label. This last filter is the most effective for perceived answer quality, provided your teams actually maintain this tagging discipline.

Synchronization frequency is configurable per source, with incremental synchronization happening automatically. Real-time mode captures changes within minutes, suiting Slack and collaborative tools where information ages quickly. Hourly mode suffices for SharePoint and Drive. Daily mode fits stable bases like Confluence, where pages rarely change multiple times a day. If the same document exists in two sources (e.g., Drive and Confluence), Enterprise Search displays only one result: deduplication compares content, not filenames.

Supported Formats

Binary files, video, audio, and executables are not indexed. Highlight OCR support to your teams: an old scanned contract becomes queryable again. This demonstration often wins over legal departments.

Long-Term Monitoring

The console displays the status of each connector: indexed documents per source, last synchronization date, indexing errors (due to corrupted files or insufficient permissions), and data volume. An error in one connector does not affect others; search continues on operational sources. While reassuring, this creates a subtle risk: no one notices a source has stopped indexing unless someone checks. Include this verification in a monthly review, along with a re-reading of the list of active connectors.

Key Takeaways

• Google Workspace, Microsoft 365, Confluence, and Slack connectors cover most needs
• Apply the principle of least privilege: index only what is necessary
• Configure filtering by folder, file type, and date to optimize result relevance
• Regularly monitor connector status via the administration console