Xyne Documentation

The project uses semantic-release, which determines the next version number based on your commit types:

• Patch version (e.g., 1.0.1 → 1.0.2): Triggered by fix commits.
• Minor version (e.g., 1.1.0 → 1.2.0): Triggered by feat commits.
• Major version (e.g., 2.0.0 → 3.0.0): Triggered by feat! or commits containing BREAKING CHANGE: in the footer.

The integration is split into two main functional areas:

1. Client Module (client.ts)

Responsible for the low-level communication with Microsoft:

• createMicrosoftGraphClient: Creates a Microsoft Graph client instance.
• makeGraphApiCall: Helper function for making API calls with retry logic.
• makePagedGraphApiCall: Helper for handling paginated requests.
• CustomAuthProvider: A custom authentication provider implementing the Microsoft Graph SDK interface.

2. Main Integration (index.ts)

Handles the high-level data processing for specific Microsoft services:

• OneDrive Files: Metadata and content indexing.
• Outlook Emails: Email content and metadata processing.
• Calendar Events: Meeting details, attendees, and attachments.
• Contacts: Contact information and organization details.

To signal a breaking change (which triggers a Major version bump in semantic-release), you must either:

1. Add a ! after the type (e.g., feat!: description).
2. Include BREAKING CHANGE: in the commit footer.

Example:

feat(api)!: update user authentication endpoints

BREAKING CHANGE: The authentication endpoints have been updated.

feat(api)!: update user authentication endpoints

BREAKING CHANGE: The authentication endpoints have been updated.
- /auth/login is now /api/v2/auth/login
- /auth/logout is now /api/v2/auth/logout

The evaluateSearchQualityLLM.ts script is a tool for quantitatively measuring the effectiveness of the Vespa search engine integration. It uses an LLM to simulate complex, conceptual user searches by generating queries based on actual document content, rather than relying on simple keyword matching.

This allows developers to:

• Benchmark search performance against LLM-generated queries.
• Regression Test to ensure changes don't degrade relevance for complex queries.
• Tune hybrid search parameters (specifically the alpha value).
• Analyze Failures by using an LLM to explain why certain queries failed to retrieve target documents.

Since Jira webhooks are sent from Atlassian's infrastructure, you can secure your endpoint by appending an authentication token as a query parameter to your webhookUrl.

1. Configure the JiraTrigger with a URL like: https://your-app.com/api/v1/webhook/jira/<id>?auth=your-secret-token.
2. In your backend webhook handler, validate the auth query parameter before processing the event.

const trigger = new JiraTrigger({
  // ... other config
  webhookUrl: 'https://your-app.com/api/v1/webhook/jira/<your-stable-webhook-id>?auth=your-secret-token',
})

// In your endpoint handler:
app.post('/webhook/jira', (req, res) => {
  if (req.query.auth !== 'your-secret-token') {
    return res.status(403).send('Unauthorized')
  }

  // Process event
})

Xyne supports two primary authentication methods for ingesting data. You only need to configure one of these types:

• OAuth Authentication: Best suited for individual users running the application for testing or personal use.
• Service Account Authentication: Best suited for organizations and production environments.

Before attempting to ingest data, ensure you have already completed the setup for your chosen authentication method.

When DEBUG_POOR_RANKINGS is set to true, the script performs an automated failure analysis to help debug search relevance issues.

1. Identification: It identifies documents that are either not found in the search results or are ranked lower than the POOR_RANK_THRESHOLD (default 10).
2. Data Collection: For these specific failures, it saves detailed debug_[HASH]_[TIMESTAMP].json files containing Vespa trace information and top result data.
3. Reporting: It aggregates these failures to identify common patterns (e.g., issues with tokenization or term weighting) and generates a search_failure_analysis.md report with actionable recommendations.

After successfully logging in with your Google account and starting the application, you can access the ingestion interface via the UI:

1. Navigate to the main application screen in your browser.
2. Locate the plug icon on the left-hand side of the interface. This icon represents the connector.
3. Clicking the connector will allow you to choose your ingestion method based on your configured authentication (either Oauth Account or Service Account).