Vector Databases

RealTimeX stores embeddings in a vector database so it can retrieve relevant context from your indexed content.

Open the system-wide storage selector from Settings > AI Providers > Vector Database.

The default experience

The default built-in option is LanceDB.

This is the simplest path for many desktop and self-hosted deployments because it runs locally and does not require a separate managed vector service.

Current database choices

The current product exposes these vector database providers:

• LanceDB
• PGVector
• Chroma
• Chroma Cloud
• Pinecone
• Zilliz Cloud
• Qdrant
• Weaviate
• Milvus
• AstraDB

Some are built for local or self-hosted use, while others are managed cloud services that require connection details and credentials.

Switching the vector database is destructive

Changing the system vector database affects the entire instance, not just one workspace.

The current product warns that switching providers will:

• reset previously embedded documents in all workspaces
• clear all embeddings from the vector database
• remove documents from workspaces
• keep uploaded source files available for re-embedding

Treat a provider switch as an infrastructure migration, not a minor preference change.

Workspace-level retrieval settings

After the system-wide database is chosen, each workspace still has its own retrieval controls.

In workspace vector settings, the current UI lets you:

• inspect the workspace vector database identifier
• inspect the current vector count
• choose Default Search or Rerank Search when the active database supports it
• set the maximum number of context snippets
• set the document similarity threshold
• reset the vector database for that workspace only

The workspace reset action is separate from the system-wide provider change. Use it when one workspace needs its vectors rebuilt without migrating the entire instance.

Search mode support

Search Mode is currently shown only for supported vector databases. In the current UI, that support is limited to databases such as LanceDB.

Use:

• Default Search for standard vector similarity lookup
• Rerank Search when you want candidate retrieval followed by reranking for better relevance

Choosing built-in vs external storage

• Use LanceDB when you want the simplest local setup.
• Use PGVector when your team already operates PostgreSQL and wants vector search there.
• Use a managed cloud service when you want remote infrastructure, external scaling, or centralized database operations.
• Use a self-hosted vector engine when you need more direct control over storage and deployment topology.

Vector database vs embedding model

The vector database stores vectors. The embedding model creates them.

If retrieval quality is poor, inspect both sides:

• Embedding Models for how content is encoded
• Vector Database settings for where vectors are stored and how workspaces retrieve them