Search index

Storage backends persist processed chunks and search across your document collection. Vector stores store chunk embeddings and find similar vectors for semantic search.

Available backends

Officially supported

Backend	Purpose
Mistral Search Index	Search storage hosted by Mistral, with no database to operate
Vespa	Vector database with schema management, ranking, clustering, and replication
Postgres	PostgreSQL with `pgvector` and `pg_textsearch` for dense and hybrid search

All built-in backends implement the VectorStoreIndex protocol, so they use the same ingestion and retrieval pipeline interfaces. Their provisioning, schema management, and query-scoping behavior differ. For details, see each backend's page.

Community

Backend	Purpose
Qdrant	Qdrant vector search engine, maintained as a third-party plugin

Community backends aren't part of the mistralai-search-toolkit package. For installation and support, see each plugin's repository.

Custom backends

Backend	Purpose
Custom vector stores	Implement your own storage backend