Vector Database

To use the Search module, you must configure a vector database. A detailed explanation of why this setup is required is provided on the Architecture page. The complete YAML configuration is documented on the Configuration page.

service
|   └── search
|   |   └── vectorDatabase
|   |   |   └── type
|   |   |   └── milvus
|   |   |   |   └── user
|   |   |   |   └── password
|   |   |   |   └── token
|   |   |   |   └── endpoint
|   |   |   |   └── consistency
|   |   |   |   └── reload
|   |   |   |   └── index
|   |   |   |   |   └── type
|   |   |   |   |   └── params
|   |   |   |   |   |   └── nlist
|   |   |   |   └── search
|   |   |   |   |   └── type
|   |   |   |   |   └── params
|   |   |   |   |   |   └── nprobe
|   |   |   └── opensearch
|   |   |   |   └── awsAuth
|   |   |   |   |   └── accessKey
|   |   |   |   |   └── enabled
|   |   |   |   |   └── region
|   |   |   |   |   └── secretKey
|   |   |   |   └── dimension
|   |   |   |   └── host
|   |   |   |   └── password
|   |   |   |   └── port
|   |   |   |   └── useSsl
|   |   |   |   └── username
|   |   |   |   └── verifyCerts
|   |   |   └── atlas
|   |   |   |   └── connectionString

The following vector database types are supported:

Milvus. A Milvus deployment typically includes the Milvus Operator, MinIO, etcd, and Pulsar components. It can run on-premises or in the cloud (for example, on Google Cloud Platform or Amazon Web Services). Trial credits are available in the AWS Ohio region, but you may choose another region if needed.
OpenSearch: Minimum supported version is 2.19.0
MongoDB Atlas

To enable a specific database type, configure the corresponding parameters in the YAML file as described below. All stored data is de-identified and represented only as vectors.

Parameter	Type	Default	Description
`vectorDatabase` → `type`	string	`milvus`, `opensearch`, `atlas`	The vector database type.
`milvus` → `user`	string	—	The user for connecting to the Milvus instance. If both user and password are provided, the system will include the corresponding header in every RPC call.
`milvus` → `password`	string	—	The password corresponding to the specified user.
`milvus` → `token`	string	—	Serves as the key for identification and authentication purposes. When a token is provided, the system adds the corresponding header to each RPC call. Users can use either user+password or token. If both are filled, the system logs in using user+password.
`milvus` → `endpoint`	string	`http://localhost:19530`	The endpoint URL for the Milvus instance.
`milvus` → `consistency`	string	`"Bounded"`	The consistency level to use when searching in a collection. Options for consistency level include `Strong`, `Bounded`, `Eventually`, `Session`, and `Customized`.
`milvus` → `reload`	boolean	false	If set to true, checks that collections for searching are in memory and loads them if not. If set to false, this option is disabled.
`milvus` → `index`	map	—	Parameters for the index to achieve better search performance.
`index` → `type`	string	`"IVF_FLAT"`	Index type, see the Milvus documentation for details.
`index` → `params` → `nlist`	integer	`128`	Number of cluster units, see the IVF_FLAT article for details.
`search` → `type`	string	`"L2"`	The similarity metric type, metrics used to measure similarity of vectors, see the Similarity Metrics article for details.
`search` → `params` → `nprobe`	integer	`5`	Number of units to query, see the IVF_FLAT article for details.
`opensearch` → `awsAuth` → `accessKey`	string	—	AWS access key ID used for authenticating requests to an OpenSearch cluster that is protected with AWS SigV4.
`opensearch` → `awsAuth` → `enabled`	boolean	`false`	Enables AWS SigV4 authentication for OpenSearch. Set to `true` if your cluster requires AWS IAM–based authentication.
`opensearch` → `awsAuth` → `region`	string	—	AWS region where the OpenSearch cluster is hosted (for example, `us-east-1`). Required when `awsAuth.enabled` is `true`.
`opensearch` → `awsAuth` → `secretKey`	string	—	AWS secret access key paired with the access key ID. Used to sign requests to OpenSearch.
`opensearch` → `dimension`	integer	`512`	Dimensionality of the vector embeddings stored in the OpenSearch k-NN index. Must match your model’s embedding size.
`opensearch` → `host`	string	`localhost`	OpenSearch host name or IP.
`opensearch` → `password`	string	`""`	Password for the specified OpenSearch user.
`opensearch` → `port`	string	`"9200"`	OpenSearch port.
`opensearch` → `useSsl`	boolean	`false`	Whether to enable HTTPS.
`opensearch` → `username`	string	`admin`	Username for OpenSearch authentication.
`opensearch` → `verifyCerts`	boolean	`false`	Whether to verify server SSL certificates when `useSsl` is `true`.
`atlas` → `connectionString`	string	—	MongoDB Atlas connection string, for example, `mongodb+srv://<user>:<password>@<cluster>/<db>?retryWrites=true&w=majority&appName=<app>`.