Docs/AI Backup/Vector DBs

Vector Database Backup

Schema-aware backup for Pinecone, Weaviate, Qdrant, Milvus, pgvector, and Chroma.

Why Back Up a Vector Database?

Vector databases store embeddings — mathematical representations of documents, images, and ideas. If your vector DB is lost or corrupted, you must re-embed everything from scratch using an embedding API. At scale, this costs thousands of dollars and takes days or weeks to complete. Backup preserves the entire index so you can restore directly in minutes.

•Cost savings: Avoid $5,000+ in re-embedding API fees.
•Time savings: Restore in minutes instead of days.
•Version control: Keep historical indexes for experimentation and rollback.
•Audit trail: Full backup history for compliance and recovery scenarios.

Supported Vector Database Providers

BackupEngine supports all major vector database platforms:

Provider	Type	Auth Method
Pinecone	Managed cloud	API key
Weaviate	Cloud or self-hosted	API key / connection string
Qdrant	Cloud or self-hosted	API key / connection string
Milvus	Self-hosted cluster	Connection string / credentials
pgvector (PostgreSQL)	Managed or self-hosted	PostgreSQL connection string
Chroma	Self-hosted or Chroma Cloud	API key / local path

How Connection Works

BackupEngine communicates with your vector database to read index data, metadata, and schema. Your credentials (API keys, connection strings) are encrypted client-side with your personal encryption key before they ever leave your device. BackupEngine servers never see plaintext API keys or database passwords.

•Client-side encryption: Credentials encrypted with AES-256-GCM using your recovery key.
•Zero-knowledge architecture: Servers orchestrate backup jobs but never access credentials.
•Secure token refresh: Long-lived tokens are rotated automatically.
•Audit logging: All backup operations are logged but never expose sensitive data.

ℹ Note

For managed services (Pinecone, Weaviate Cloud), use read-only API keys with minimal permissions if available. This limits risk if a credential were ever exposed.

Setup Steps

Connect a vector database in three steps:

•Open customer portal → AI Assets → Vector Databases.
•Click 'Connect Database' and choose your provider.
•Paste your API key or connection string, then select which indexes to back up.

CLI Examples

Vector database backup and restore commands

# Connect a new vector database (Pinecone)
backupengine ai-assets connect --provider pinecone

# List all connected vector databases
backupengine ai-assets list --type ai_vector_db

# Start a backup of a specific vector database
backupengine ai-assets backup --type ai_vector_db --provider pinecone

# List backup snapshots
backupengine ai-assets snapshots --asset "customer-rag-index"

# Restore to a specific point-in-time
backupengine ai-assets restore \
  --asset "customer-rag-index" \
  --version latest \
  --target-instance "qdrant-prod-02"

Restore & Recovery

Vector database backups support multiple restore scenarios:

•In-place restore: Overwrite the current index with a backup version (careful: this replaces all data).
•Restore to new index: Create a new index from a backup and test before switching traffic.
•Cross-provider restore: Back up from Pinecone, restore to Qdrant (schema is translated automatically).
•Selective restore: Restore only specific vector namespaces or collections, not the entire index.

⚠ Warning

In-place restore overwrites all current data. Always test restores in a non-production environment first. For critical databases, restore to a new index and verify before switching applications over.

ML Model Registry Backup AI Workspace Backup