Glean Data Handling

Glean Data Handling

Overview

Glean enterprise search ingests documents from dozens of connectors (Google Drive, Confluence, Slack, Jira, Salesforce, etc.) and builds a unified search index with permission-aware access control. Data types include indexed document content, connector metadata, user permission maps, query logs, and search analytics. All document content must be PII-filtered before indexing, permission boundaries must be preserved to prevent data leakage across teams, and retention policies must be enforced to comply with corporate governance and GDPR/CCPA obligations.

Data Classification

Data Import

interface GleanDocument {
  id: string; datasource: string; title: string;
  body: string; permissions: { allowedUsers?: string[]; allowAnonymousAccess?: boolean };
  updatedAt: string; url: string;
}

async function indexDocuments(docs: GleanDocument[], datasource: string) {
  // PII strip before indexing
  const sanitized = docs.map(doc => ({
    ...doc,
    body: stripPII(doc.body),
  }));
  // Batch upload with pagination (max 100 per request)
  for (let i = 0; i < sanitized.length; i += 100) {
    const batch = sanitized.slice(i, i + 100);
    await fetch(`https://customer-be.glean.com/api/index/v1/bulkindexdocuments`, {
      method: 'POST',
      headers: { Authorization: `Bearer ${process.env.GLEAN_INDEXING_TOKEN}`, 'Content-Type': 'application/json' },
      body: JSON.stringify({ datasource, documents: batch }),
    });
  }
}

function stripPII(text: string): string {
  return text
    .replace(/\b[\w.+-]+@[\w-]+\.[\w.]+\b/g, '[EMAIL_REDACTED]')
    .replace(/\b\d{3}[-.]?\d{3}[-.]?\d{4}\b/g, '[PHONE_REDACTED]')
    .replace(/\b\d{3}-\d{2}-\d{4}\b/g, '[SSN_REDACTED]');
}

Data Export

async function exportSearchAnalytics(startDate: string, endDate: string) {
  const res = await fetch(`https://customer-be.glean.com/api/v1/analytics`, {
    method: 'POST',
    headers: { Authorization: `Bearer ${process.env.GLEAN_API_TOKEN}`, 'Content-Type': 'application/json' },
    body: JSON.stringify({ startDate, endDate, metrics: ['query_count', 'click_through', 'zero_results'] }),
  });
  const data = await res.json();
  // Redact user identifiers from analytics export
  return data.results.map((r: any) => ({ ...r, userId: undefined, query: r.query?.length > 3 ? r.query : '[SHORT_QUERY_REDACTED]' }));
}

Data Validation

function validateDocument(doc: GleanDocument): string[] {
  const errors: string[] = [];
  if (!doc.id || doc.id.length > 512) errors.push('Invalid document ID');
  if (!doc.datasource) errors.push('Missing datasource identifier');
  if (!doc.title || doc.title.length > 1000) errors.push('Title missing or exceeds 1000 chars');
  if (!doc.body || doc.body.length === 0) errors.push('Empty document body');
  if (!doc.permissions) errors.push('Missing permissions — defaults to deny-all');
  if (doc.updatedAt && isNaN(Date.parse(doc.updatedAt))) errors.push('Invalid updatedAt timestamp');
  return errors;
}

Compliance

• PII stripped from document body before indexing (emails, phones, SSNs)
• Permission boundaries enforced: allowedUsers scope matches source system ACLs
• Connector credentials stored in secret manager, rotated quarterly
• Search query logs retained max 90 days, purged via automated job
• GDPR right-to-erasure: delete all indexed content referencing a specific user on request
• CCPA: honor do-not-sell signals for search analytics data
• SOC 2 Type II audit trail for all indexing and deletion operations

Error Handling

Prerequisites

• A documented data owner and approved classification/retention policy for the source being indexed.
• A non-production sample containing fictitious identities, plus an allowlisted destination and credentials scoped only to that datasource.
• A reviewed redaction and deletion plan; pattern matching is a safeguard, not proof that sensitive data is absent.

Instructions

1. Inventory fields and source ACLs before indexing; default any unknown permission or classification to deny and quarantine the record.
2. Run redaction and schema validation on a bounded staging batch, logging only record counts, rule versions, and opaque document identifiers.
3. Compare source and proposed index ACLs for both an allowed and denied test principal before approving a production batch.
4. Send idempotent, bounded batches to the approved datasource and retain a reversible manifest of submitted opaque IDs and timestamps.
5. On a failed validation, stop the batch, delete only the staged artifacts, and escalate to the data owner rather than weakening redaction or ACL rules.

Output

Return a handling receipt with source classification, approved destination, policy and redaction-rule versions, input/accepted/quarantined counts, ACL comparison result, retention date, and deletion or rollback reference. Do not emit bodies, queries, emails, phone numbers, or access tokens.

Examples

Example receipt: source=staging-wiki; classification=confidential; accepted=96; quarantined=4; acl_probe=allow+deny pass; retention=2026-10-01; redaction_rules=v3. The sample is fictional and identifies no employee or document.

Resources

• Glean Indexing API
• Glean Search API

Next Steps

See glean-security-basics.