Do I need multi-tenant search isolation?

Maybe — but the right answer depends on your data sensitivity, tenant size, and compliance requirements.

You probably do need search isolation if:

• Tenants must never see each other’s data, even by mistake.
• You have regulated data (HIPAA, SOC 2 controls, GDPR concerns, financial data, etc.).
• You need strong tenant-level access control and auditability.
• One tenant’s query load could impact others and you need performance isolation.
• Tenants can customize analyzers, schemas, relevance rules, or embeddings in ways that shouldn’t affect others.

You may not need it if:

• Data is low-risk and tenants are logically separated well enough by filters.
• You have a small number of tenants and simple access patterns.
• You’re okay with a shared index plus strict tenant_id filtering.
• Cost/operational simplicity matters more than hard isolation.

Common isolation levels

1. Shared index, filter by tenant_id

   • Cheapest and simplest
   • Good for many SaaS apps
   • Weakest isolation; relies on correct application logic

2. Separate namespace / collection per tenant

   • Better logical separation
   • Easier deletion and lifecycle management
   • Still shared infrastructure

3. Separate index per tenant

   • Stronger isolation
   • Better for large or sensitive tenants
   • More operational overhead

4. Separate cluster / account / region

   • Strongest isolation
   • Best for enterprise or compliance-heavy cases
   • Highest cost/complexity

A practical rule of thumb

Ask yourself:

• What happens if a query bug leaks one tenant’s result to another?
• Do customers or auditors expect hard isolation?
• Do tenants need different retention, encryption, or deletion policies?
• How many tenants do you expect, and how large are they?

If the answer to any of the first three is “bad,” you likely want stronger isolation than just a filter.

If you want, I can help you choose the right model based on:

• number of tenants,
• data sensitivity,
• search volume,
• and your current search architecture.