Connect for On-Demand Search

Use a project endpoint when you want to run on-demand search or query workloads with compute from an on-demand cluster.

Note

This page is for connecting to a project endpoint for on-demand search. If you want to connect to a Free, Serverless, or Dedicated serving cluster, see Connect to Serving Clusters.

Endpoint format

Endpoint type	Endpoint pattern	Use for
Project endpoint	https://{project-id}.{region}.api.zillizcloud.com	Data import, batch search, query, get, search, and hybrid search through an on-demand cluster.

Before you begin

• Get the project endpoint from the Zilliz Cloud console.

• Get the on-demand cluster ID that should provide compute resources for the search workload.

• Create an API key with sufficient permissions for the project and target data.

• Install a Milvus SDK for your use case. For details, refer to Install SDKs.

Connect to a Project Endpoint

Create a MilvusClient with the project endpoint and specify the on-demand cluster that should serve the request.

python

from pymilvus import MilvusClient

client = MilvusClient(
    uri="https://{project-id}.{region}.api.zillizcloud.com",
    cluster="inxx-xxxxxxxxxxxxxxx",
    token="YOUR_API_KEY",
)

Create a Search Session

Use a session object to attach your operations to the on-demand cluster.

python

session = client.session(cluster_id="inxx-xxxxxxxxxxxxxxx")

Then use the session to run DQL operations such as query, get, search, and hybrid_search.

python

results = session.search(
    collection_name="my_collection",
    data=[[0.1, 0.2, 0.3, 0.4]],
    anns_field="vector",
    limit=10,
)

print(results)

Authentication

When connecting to a project endpoint, use a valid API key as the authentication token.

Cluster credentials in username:password format are for serving cluster endpoints. For on-demand search through a project endpoint, use an API key with the required project permissions.

When to Use This Connection

Use the project endpoint for batch processing, exploration, validation, experiments, and other workloads where on-demand compute is a better fit than always-on serving.

For production applications that require the full Collection API and always-on low-latency serving, connect to a Free, Serverless, or Dedicated serving cluster instead. See Connect to Serving Clusters for the serving cluster endpoint formats and connection examples.