Skip to main content

Create a Data Pipeline with Milvus/Zilliz Cloud

This quickstart will walk you through creating a pipeline that prepares your data for AI agents. You'll set up a pipeline that transforms content from the Vectorize documentation into structured, searchable context in Milvus - giving agents the foundation they need to reason over your data, not just retrieve it.

Milvus is the underlying vector database; Zilliz Cloud is the fully managed service of Milvus.

Before you begin

Before starting, ensure you have access to the credentials, connection parameters, and API keys as appropriate for the following:

Step 1: Create a Zilliz Cloud Database

These instructions show how to create a cluster and database on Zilliz' free plan. A cluster is a managed instance of Milvus.

  1. Log in to Zilliz, and select Clusters in the menu.

  2. Select Create Free Cluster.

    Create Free Cluster

  3. Choose "Free," name your cluster, select your cloud region, then click Create.

    Configure and Create Cluster

  4. Save and securely store your username and password.

    Save Username and Password

  5. Your cluster will be created.

    Cluster Creation

  6. Once your cluster has been created, it'll show up as Running.

    Running Cluster

Step 2: Create a data pipeline on Vectorize

Create a New Data Pipeline

To configure a vector database integration to connect to your Zilliz Cloud instance:

  1. Click Vector Databases from the main menu.

  2. Click New Vector Database Integration from the Vector Databases page.

  3. Select the Milvus card.

    Milvus Card

  4. Enter the parameters in the form using the Milvus Parameters table below as a guide, then click Create Milvus Integration.

    Create Milvus Integration

Milvus Parameters

FieldDescriptionRequired
NameA descriptive name to identify the integration within Vectorize.Yes
Public EndpointThe public endpoint for your cluster.Yes
TokenThe cluster's token.Yes, unless you provide a username/password
UsernameThe cluster's username.Yes, unless you provide a token
PasswordThe cluster's password.Yes, unless you provide a token

When you specify your Milvus integration in your pipeline configuration, Vectorize writes vector data to your Milvus instance.

Configuring the Milvus integration in a data pipeline

You can think of the Milvus integration as having two parts to it. The first is authorization with your Milvus cluster. This part is re-usable across pipelines and allows you to connect to this same application in different pipelines without providing the credentials every time.

The second part is the configuration that's specific to your data pipeline. This is where you specify the name of the collection in your Milvus cluster. If the collection does not already exist, Vectorize will create it for you.

Create Milvus Integration

Configure AI Platform

  1. Click on + New AI Platform.

    New AI Platform

  2. Select OpenAI from the AI platform options.

    Select OpenAI

  3. In the OpenAI configuration screen:

    • Enter a descriptive name for your OpenAI integration.
    • Enter your OpenAI API Key.

    Configure OpenAI

  4. Leave the default values for embedding model, chunk size, and chunk overlap for the quickstart.

    Set Embedding Model

Add Source Connectors

  1. Click on Add Source Connector.

Web Crawler Source

  1. Choose the type of source connector you'd like to use. In this example, select Web Crawler.

Choose Web Crawler

Configure Web Crawler Integration

  1. Name your web crawler source connector, e.g., vectorize-docs.
  2. Set Seed URL(s) to https://docs.vectorize.io.

Configure Web Crawler

  1. Click Create Web Crawler Integration to proceed.

Configure Web Crawler Pipeline

  1. Accept all the default values for the web crawler pipeline configuration:
    • Throttle Wait Between Requests: 500 ms
    • Maximum Error Count: 5
    • Maximum URLs: 1000
    • Maximum Depth: 50
    • Reindex Interval: 3600 seconds

Web Crawler Pipeline Configuration

  1. Click Save Configuration.

Verify Source Connector and Schedule Pipeline

  1. Verify that your web crawler connector is visible under Source Connectors.
  2. Click Next: Schedule RAG Pipeline to continue.

Verify Source Connector

Schedule Data Pipeline

  1. Accept the default schedule configuration
  2. Click Create RAG Pipeline.

Schedule RAG Pipeline

Step 3: Monitor and Test Your Pipeline

Monitor Pipeline Creation and Backfilling

  1. The system will now create, deploy, and backfill the pipeline.
  2. You can monitor the status changes from Creating Pipeline to Deploying Pipeline and Starting Backfilling Process.

Pipeline Creation

  1. Once the initial population is complete, the data pipeline will begin crawling the Vectorize docs and writing vectors to your Milvus index.

Pipeline Backfilling

View Data Pipeline Status

  1. Once the website crawling is complete, your data pipeline will switch to the Listening state, where it will stay until more updates are available.

Pipeline Listening State

Step 4: Test Your Pipeline in the RAG Sandbox

Access the RAG Sandbox

  1. From the main pipeline overview, click on the RAG Pipelines menu item to view your active pipelines.

Open RAG Pipeline Menu

  1. Find your pipeline in the list of pipelines.
  2. Click on the magnifying glass icon under the RAG Sandbox column to open the sandbox for your selected pipeline.

Open RAG Sandbox

Query Your Data

  1. In the sandbox, you can ask questions about the data you've ingested.
  2. Type a question related to your dataset in the Question field. For example, "What is Vectorize?" since you're working with the Vectorize documentation.
  3. Click Submit to send the question.

Ask a Question

Review Results

  1. After submitting your question, the sandbox will retrieve relevant chunks from your vector database and display them in the Retrieved Context section.
  2. The response from the language model (LLM) will be displayed in the LLM Response section.
    • The Retrieved Context section shows the chunks that were matched with your question.
    • The LLM Response section provides the final output based on the retrieved chunks.

Retrieved Chunks and LLM Response

  1. You can continue to ask different questions or refine your queries to explore your dataset further.
  2. The sandbox allows for dynamic interactions with the data stored in your vector database.

That's it! You've successfully created a data pipeline that transforms your content into structured context, ready for AI agents to reason over and make intelligent decisions.

Was this page helpful?