vai index

Create, list, and delete MongoDB Atlas Vector Search indexes.

Synopsis

vai index create --db <database> --collection <name> --field <name> [options]
vai index list --db <database> --collection <name> [options]
vai index delete --db <database> --collection <name> --index-name <name> [options]

Description

vai index manages vector search indexes on MongoDB Atlas collections. Indexes are required for $vectorSearch queries to work. The command has three subcommands: create, list, and delete.

---

vai index create

Create a new vector search index on a collection.

Flag	Description	Default
--db <database>	Database name (required)	—
--collection <name>	Collection name (required)	—
--field <name>	Embedding field name (required)	—
-d, --dimensions <n>	Vector dimensions	1024
-s, --similarity <type>	Similarity function: cosine, dotProduct, euclidean	cosine
-n, --index-name <name>	Index name	vector_index
--json	Machine-readable JSON output	—
-q, --quiet	Suppress non-essential output	—

vai index create --db myapp --collection docs --field embedding --dimensions 1024

note

Index creation is asynchronous. It may take a few minutes for the index to become ready for queries.

---

vai index list

List all search indexes on a collection.

Flag	Description	Default
--db <database>	Database name (required)	—
--collection <name>	Collection name (required)	—
--json	Machine-readable JSON output	—
-q, --quiet	Suppress non-essential output	—

vai index list --db myapp --collection docs

---

vai index delete

Drop a search index by name.

Flag	Description	Default
--db <database>	Database name (required)	—
--collection <name>	Collection name (required)	—
-n, --index-name <name>	Index name to delete (required)	—
--json	Machine-readable JSON output	—
-q, --quiet	Suppress non-essential output	—

vai index delete --db myapp --collection docs --index-name vector_index

Examples

Create an index with dot product similarity

vai index create --db myapp --collection docs --field embedding \
  --dimensions 1024 --similarity dotProduct --index-name my_index

List indexes as JSON

vai index list --db myapp --collection docs --json

Tips

• You need at least one vector search index before using vai search or vai query.
• Use --similarity cosine (the default) unless you have specific requirements for dot product or Euclidean distance.
• The vai pipeline --create-index flag can auto-create an index as part of the pipeline.
• If an index with the same name already exists, create will fail — use a different --index-name or delete the existing one first.

Related Commands

• vai search — Search using a vector index
• vai pipeline — Create indexes automatically with --create-index