> ## Documentation Index
> Fetch the complete documentation index at: https://docs.paradedb.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Indexing Vectors

> Vectors live alongside text and filters in the ParadeDB index

<Note>
  This is a beta feature available in versions `0.25.0` and above. See [How
  Vector Search Works](/documentation/vector/overview) for background.
</Note>

<Note>
  Make sure the [pgvector](https://github.com/pgvector/pgvector) extension is
  installed first. ParadeDB uses pgvector's vector types, but not its HNSW or
  IVF indexes.
</Note>

The ParadeDB index can index pgvector's `vector` type alongside your text and other columns. This lets you combine vector search with full text search and filters in a single index,
which can significantly improve latency/recall for selective queries.

## Create the Index

The `mock_items` table comes with an `embedding` column of type `vector(8)` populated with sample embeddings.
In this example, `embedding` is added to the ParadeDB index with cosine similarity as the distance function.

<CodeGroup>
  ```sql SQL theme={null}
  CREATE INDEX search_idx ON mock_items
  USING paradedb (id, description, category, embedding vector_cosine_ops)
  WITH (key_field='id');
  ```

  ```ts Drizzle theme={null}
  // Vector search support for Drizzle is coming very soon.
  ```

  ```python Django theme={null}
  # Vector search support for Django is coming very soon.
  ```

  ```python SQLAlchemy theme={null}
  # Vector search support for SQLAlchemy is coming very soon.
  ```

  ```ruby Rails theme={null}
  # Vector search support for Rails is coming very soon.
  ```

  ```cs EF Core theme={null}
  // Vector search support for EF Core is coming very soon.
  ```
</CodeGroup>

The desired distance function is encoded into the index definition, and cannot be changed without reindexing.
Use `vector_l2_ops` for L2 distance, `vector_ip_ops` for inner product, or `vector_cosine_ops` for cosine distance.

<Note>
  Only pgvector's `vector` type is supported. The `halfvec`, `sparsevec`, and
  `bit` types are not yet indexable.
</Note>

<Note>
  If you track index build progress with
  [`pg_stat_progress_create_index`](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING),
  you may notice progress appear to "stop" at intervals. This is expected:
  vectors are clustered with k-means at these points, which is computationally
  expensive.
</Note>

## Index Options

ParadeDB uses a SPANN-style vector index, which is similar to an IVF index but with additional structures to improve
recall and latency, especially over large datasets. The following `WITH` options control how vectors are clustered and indexed. All are set at index build time and apply to every vector field in the index. An example:

<CodeGroup>
  ```sql SQL theme={null}
  CREATE INDEX search_idx ON mock_items
  USING paradedb (id, description, category, embedding vector_cosine_ops)
  WITH (key_field='id', centroid_ratio=0.01, training_samples_per_centroid=32, cluster_replication=1);
  ```

  ```ts Drizzle theme={null}
  // Vector search support for Drizzle is coming very soon.
  ```

  ```python Django theme={null}
  # Vector search support for Django is coming very soon.
  ```

  ```python SQLAlchemy theme={null}
  # Vector search support for SQLAlchemy is coming very soon.
  ```

  ```ruby Rails theme={null}
  # Vector search support for Rails is coming very soon.
  ```

  ```cs EF Core theme={null}
  // Vector search support for EF Core is coming very soon.
  ```
</CodeGroup>

<ParamField body="centroid_ratio" default={0.01}>
  Vectors are clustered by proximity, and a centroid is a cluster's
  representative vector. This setting controls the number of centroids to build,
  as a fraction of the number of indexed vectors (`num_centroids =
      centroid_ratio * num_vectors`). More centroids produce smaller clusters,
  improving recall at the cost of a slower, more memory-intensive build. Must be
  between `0.000001` and `1.0`.
</ParamField>

<ParamField body="training_samples_per_centroid" default={32}>
  The number of vectors sampled per centroid to train the k-means clustering at
  build time. Higher values yield better-quality centroids at the cost of a
  slower build. Must be between `1` and `100000`.
</ParamField>

<ParamField body="cluster_replication" default={1}>
  The number of clusters each vector is written into: its primary cluster plus
  up to `cluster_replication - 1` next-nearest clusters. Higher values can
  improve recall for filtered queries at the cost of a larger index. The default
  of `1` disables replication.
</ParamField>