Vector Stores

Vectorstores

PrivateGPT supports Qdrant, Milvus, Chroma, PGVector and ClickHouse as vectorstore providers. Qdrant being the default.

In order to select one or the other, set the vectorstore.database property in the settings.yaml file to qdrant, milvus, chroma, postgres and clickhouse.

1
vectorstore:
2
  database: qdrant

Qdrant configuration

To enable Qdrant, set the vectorstore.database property in the settings.yaml file to qdrant.

Qdrant settings can be configured by setting values to the qdrant property in the settings.yaml file.

The available configuration options are:

By default Qdrant tries to connect to an instance of Qdrant server at http://localhost:3000.

To obtain a local setup (disk-based database) without running a Qdrant server, configure the qdrant.path value in settings.yaml:

1
qdrant:
2
  path: local_data/private_gpt/qdrant

Milvus configuration

To enable Milvus, set the vectorstore.database property in the settings.yaml file to milvus and install the milvus extra.

$
poetry install --extras vector-stores-milvus

The available configuration options are:

To obtain a local setup (disk-based database) without running a Milvus server, configure the uri value in settings.yaml, to store in local_data/private_gpt/milvus/milvus_local.db.

Chroma configuration

To enable Chroma, set the vectorstore.database property in the settings.yaml file to chroma and install the chroma extra.

$
poetry install --extras chroma

By default chroma will use a disk-based database stored in local_data_path / “chroma_db” (being local_data_path defined in settings.yaml)

PGVector

To use the PGVector store a postgreSQL database with the PGVector extension must be used.

To enable PGVector, set the vectorstore.database property in the settings.yaml file to postgres and install the vector-stores-postgres extra.

$
poetry install --extras vector-stores-postgres

PGVector settings can be configured by setting values to the postgres property in the settings.yaml file.

The available configuration options are:

For example:

1
vectorstore:
2
  database: postgres
3
4
postgres:
5
  host: localhost
6
  port: 5432
7
  database: postgres
8
  user: postgres
9
  password: <PASSWORD>
10
  schema_name: private_gpt

The following table will be created in the database

postgres=# \d private_gpt.data_embeddings
                                      Table "private_gpt.data_embeddings"
  Column   |       Type        | Collation | Nullable |                         Default
-----------+-------------------+-----------+----------+---------------------------------------------------------
 id        | bigint            |           | not null | nextval('private_gpt.data_embeddings_id_seq'::regclass)
 text      | character varying |           | not null |
 metadata_ | json              |           |          |
 node_id   | character varying |           |          |
 embedding | vector(768)       |           |          |
Indexes:
    "data_embeddings_pkey" PRIMARY KEY, btree (id)
postgres=# 

The dimensions of the embeddings columns will be set based on the embedding.embed_dim value. If the embedding model changes this table may need to be dropped and recreated to avoid a dimension mismatch.

ClickHouse

To utilize ClickHouse as the vector store, a ClickHouse database must be employed.

To enable ClickHouse, set the vectorstore.database property in the settings.yaml file to clickhouse and install the vector-stores-clickhouse extra.

$
poetry install --extras vector-stores-clickhouse

ClickHouse settings can be configured by setting values to the clickhouse property in the settings.yaml file.

The available configuration options are:

For example:

1
vectorstore:
2
  database: clickhouse
3
4
clickhouse:
5
  host: localhost
6
  port: 8443
7
  username: admin
8
  password: <PASSWORD>
9
  database: embeddings
10
  secure: false

The following table will be created in the database:

clickhouse-client
:) \d embeddings.llama_index
                                   Table "llama_index"
  № |  name     | type                                         | default_type | default_expression | comment | codec_expression | ttl_expression
----|-----------|----------------------------------------------|--------------|--------------------|---------|------------------|---------------
  1 | id        | String                                       |              |                    |         |                  |
  2 | doc_id    | String                                       |              |                    |         |                  |
  3 | text      | String                                       |              |                    |         |                  |
  4 | vector    | Array(Float32)                               |              |                    |         |                  |
  5 | node_info | Tuple(start Nullable(UInt64), end Nullable(UInt64)) |       |                    |         |                  |
  6 | metadata  | String                                       |              |                    |         |                  |
clickhouse-client

The dimensions of the embeddings columns will be set based on the embedding.embed_dim value. If the embedding model changes, this table may need to be dropped and recreated to avoid a dimension mismatch.