Please use the form below to provide your feedback. Because your feedback is valuable to us, the information you submit in this form is recorded in our issue tracking system (JIRA), which is publicly available. You can track the status of your feedback using the ticket number displayed in the dialog once you submit the form.

Vectorize Structured Data from Amazon S3

  • how-to
Use a Structured Data Workflow to automatically generate embedding vectors from JSON data in an Amazon S3 Bucket. Use embedding vectors for similarity searches on your data.

The Vectorization Service automatically creates a Vector Search index for your embeddings - letting you get started right away with Vector Search. You can use Vector Search to support Retrieval Augmented Generation (RAG) in your applications, or for other vector similarity use cases.

Your data must already be extracted, filtered, and chunked in preparation for generating embeddings. Data must be in JSON, JSON List, or JSON Lines format. If your data is not already in JSON format, see Process and Vectorize Unstructured Data.

To generate your embeddings, you can use a model hosted by the AI Data Plane Model Service, OpenAI, or Amazon Bedrock. The AI Data Plane stores the generated vector embeddings and Vector Search index in an operational cluster.

Prerequisites

  • You have read-only credentials available for the Amazon S3 bucket where your data is stored. For more information about AWS access keys, see the AWS documentation.

  • If you want to use a model hosted on the AI Data Plane, you must have:

  • If you want to use a model hosted by OpenAI, you have your OpenAI API Key. For more information about how to find your OpenAI API Key, see the OpenAI Help Center.

  • If you want to use an Amazon Bedrock model, you need a short-term Amazon Bedrock API key. For more information, see Add an Amazon Bedrock API Key.

  • You have deployed an operational cluster that has the following:

    • Couchbase Server version 8.0 or later.

    • The Search Service and Eventing Service running on at least 1 Service Group. For more information, see Services and Service Groups.

    • A bucket that can store the Vector Search index and any generated vector embeddings.

      Use any bucket settings you would prefer for your particular use case. For more information, see Manage Buckets.

Procedure

To create a new Structured Data workflow and process your JSON data:

  1. Go to AI Data Plane  Workflows.

  2. Click Create New Workflow.

  3. Click Structured Data from External sources.

  4. In the Workflow Name field, enter a name to identify your Structured Data Workflow, or accept the automatically generated name.

    Workflow names can be a maximum of 128 characters and can include letters (A-Z, a-z), numbers (0-9), dashes (-), and underscores (_).

  5. Click Setup Workflow.

  6. Configure Your Amazon S3 Bucket.

  7. Do 1 of the following:

    1. To use an existing field in your data for your document ID values, click Manually Enter Document ID Field and provide your document ID field name.

    2. To automatically generate document IDs in a new field for all documents, click Autogenerate ID Field.

  8. Configure Your Source Fields.

  9. Choose whether to Create HyperScale Vector Index (now) or Create HyperScale Vector Index (later).

  10. Under Destination Cluster, in the Destination Operational Cluster list, select the cluster you configured in the Prerequisites.

  11. Set the Destination Bucket, Destination Scope, and Destination Collection for your vector embeddings.

  12. Choose Your Embedding Model.

  13. Verify your workflow configuration.

  14. Click Run Workflow.

Do not delete or modify the metadata scope, collections, or Eventing functions created by your new Workflow. If you modify or delete the metadata or functions, you must delete your Workflow and create a new one.

Configure Your Amazon S3 Bucket

Choose whether to use a new Amazon S3 bucket or choose an S3 bucket that you have already saved as an integration with the Couchbase AI Data Plane.

You can manage your saved Amazon S3 bucket credentials from the Integrations page.

  • New Amazon S3 Bucket

  • Use Existing Amazon S3 Bucket

To configure a new Amazon S3 bucket:

  1. Click Add New S3 Bucket Integration.

  2. In the Integration Name field, enter a name to use to identify your credentials and make it easier to manage them from the Integrations page.

  3. Enter the details and credentials for accessing your Amazon S3 bucket.

    It’s recommended to use read-only credentials for your S3 bucket. Make sure you have your Access Key ID and its Secret Access Key.

    You can also choose to use temporary credentials, supported by a session token. For more information about configuring temporary credentials and session tokens, see the AWS documentation.

  4. Click Add Credentials.

  5. In the S3 Bucket Integration list, select your new S3 bucket.

  6. In the Select type of files list, select the specific format for your JSON data.

  7. Verify your S3 Integration Summary.

  8. Continue with the rest of the Procedure.

To use an existing Amazon S3 bucket that you added to the AI Data Plane:

  1. In the S3 Bucket Integration list, select the S3 bucket where your structured data is stored.

  2. In the Select type of files list, select the specific format for your JSON data.

  3. Verify your S3 Integration Summary.

  4. Continue with the rest of the Procedure.

Configure Your Source Fields

You must configure whether the Vectorization Service should store all vectors generated from your documents in a single field, or create a custom source field mapping.

  • Map all source fields to a single vector field

  • Create custom source field mappings

To map all of the fields in your documents to a single vector field:

  1. Click Map all source fields to a single vector field.

  2. (Optional) Under Vector Field, enter a name for the field where you want to store your vectors.

  3. Continue with the rest of the Procedure.

To create custom mappings and only vectorize specific fields from your documents:

  1. Click Create custom source field mappings.

  2. Under Source Fields, click the list.

  3. Select every field from your source documents that you want to vectorize and store in a single field.

  4. In the corresponding field under Vector Field, enter a name for the new vector field for your selected source field or fields.

  5. (Optional) To map additional fields to another vector field, click Add more mapping and repeat Steps 2-4.

  6. Continue with the rest of the Procedure.

Choose Your Embedding Model

You can choose to use an embedding model hosted by the AI Data Plane Model Service, hosted by OpenAI, or hosted by Amazon Bedrock to vectorize your data.

  • Use an AI Data Plane Model

  • Use an OpenAI Model

  • Use an AWS Bedrock Model

To use an AI Data Plane Model:

  1. Select Capella Model as your model source.

  2. Select the specific Capella embedding model you want to use in this workflow.

  3. Upload or manually enter your embedding model’s API Key ID and API Key Token. For more information about API keys for AI Data Plane models, see Get Started with the Couchbase AI Data Plane APIs.

  4. (Optional) Choose whether to set up Private Networking for your AI Data Plane embedding model. For more information about Private Networking for the Couchbase AI Data Plane, see Add an AWS PrivateLink Connection.

  5. Click Next.

  6. Continue with the rest of the Procedure.

To use an OpenAI model:

  1. Select OpenAI as your model source.

  2. Select the specific OpenAI model you want to use in this workflow.

  3. (Optional) To use a new OpenAI API key, click Add New OpenAI API Key.

    1. Enter a name to identify your API Key in Capella.

    2. Enter your Secret Access Key from OpenAI.

    3. Click Add Key.

  4. Select the OpenAI API Key you want to use.

  5. Click Next.

  6. Continue with the rest of the Procedure.

    Workflows do not use the OpenAI Batch API.

Before you begin, generate a short-term Amazon Bedrock API key and add it to the Couchbase AI Data Plane integrations. For more information, see Add an Amazon Bedrock API Key.

To use an AWS Bedrock model:

  1. Select AWS Bedrock as your model source.

  2. Select the Bedrock Model type you want to use in this workflow.

  3. Select the model’s Dimensions.

  4. (Optional) To use a new Amazon Bedrock credential, click Add New Bedrock Credentials.

    1. Enter a name to identify your key in the AI Data Plane.

    2. Select the AWS region where you generated your short-term key.

    3. Enter your short-term Amazon Bedrock API key. The key must start with bedrock-api-key-.

    4. Click Add Credentials.

  5. Select the Bedrock credential you want to use.

  6. (Optional) Choose whether to set up private networking for your AWS Bedrock model. To enable private networking, your cluster and credential must be in the same region.

    • You can change the AWS Region for an Amazon Bedrock credential only if it’s not associated with an active workflow that has private networking enabled. For more information, see Edit an API Key or Credential.

    • Workflows that use the same cluster reuse the same private endpoint automatically. Other clusters, even in the same region, need their own private endpoint.

    If you do not set up private networking, your data traverses the public Internet with TLS encryption. You cannot enable private networking after creating a workflow; you must delete the original workflow and recreate it with private networking enabled.

    Capella bills your account for AWS PrivateLink when you have private endpoints enabled. For more information, see Private Endpoints Billing.

  7. Click Next.

  8. Continue with the rest of the Procedure.

Next Steps

The AI Data Plane UI shows the documents that have been processed by your Structured Data Workflow. You can click the Failed icon to view error information for failed documents.

Structured Data Workflows display with an S3 - Structured Type.

For more information about Workflow statuses, see Workflow Statuses.

You can also: