Ingest your content

Edit online

Upload content from a remote S3‑compatible location and ingest it into a dedicated collection source. Then use the Chat interface to run queries and get relevant responses based on the ingested documents.

About this task

Note: The following steps guide you through ingesting content using the watsonx Assistant for the Z Management Console. You can also ingest content through the CLI. For detailed instructions, see Ingesting content through Command Line Interface (CLI). In Step 1, replace <CI_URL> with the client ingestion endpoint in the format https://<hostname>:8443/client-ingestion/, where <hostname> is the fully qualified hostname of the system on which the watsonx Assistant for Z Trial is deployed (obtain this value by running hostname -f on the host machine).

Procedure

  1. Log in to the watsonx Assistant for Z Management Console
  2. Click Content Ingestion, and click Ingest Data.
    The Ingest Content modal appears.
  3. Provide the following information:
    Remote source type
    1. Select the source type, then click Next.
    Credentials
    1. URL: URL of the remote source from which the content must be ingested.
    2. Access Key ID: Access key of the remote source.
    3. Secret Access Key: Secret key of the remote source.
    4. Bucket Name: Name of the bucket containing the source files.
      Note: Supported file formats include: PDF, DOCX, CSV, and Markdown.
    Collection Source
    1. Collection Source: Choose a collection source where you want to ingest your content. You can either select one from the available list or create a new collection source.
      To create a collection source:
      1. Select New collection source .
      2. Enter a name in the Collection name field.
    2. Tabular Support: Stores data in a structured format, enabling the LLM to retrieve information efficiently (enabled by default).
    3. PII Check: Prevents the ingestion of personally identifiable information (PII).
      The following PII checks are currently supported:
      • Financial identifiers: Bank account numbers and credit card numbers, including Amex, Diners, Discover, JCB, Mastercard, Visa, and other types
      • Government identifiers: Social Security Numbers (SSN)
      • Contact information: Email addresses, phone numbers, and credentials such as access keys (restricted to bearer tokens only)
      • IP addresses
      • MAC addresses
    4. OCR: Processes and extracts images from the content source.
    5. Click Ingest to begin processing.
    Data in the processing stage appears under your defined collection source name from the specified bucket.
    Once the ingestion completes, the collection source appears on the Content Ingestion page. You can:
    • Search for a collection source by name using the search icon.
    • Filter collection sources by type or ingestion status by using the filter icon.
    • Delete collection sources by selecting them and clicking Delete.

    Languages supported by the Granite and gpt-oss-120b models:

    Multiple language configuration allows the zRAG agent to translate both user inputs and responses, managing translation dynamically based on the user’s requested language.
    Language Granite gpt-oss-120b
    Chinese ✓ ✓
    Czech ✓ ✓
    English ✓ ✓
    French ✓ ✓
    German ✓ ✓
    Italian ✓ ✓
    Japanese ✓ ✓
    Korean ✓ ✓
    Portuguese ✓ ✓
    Spanish ✓ ✓
    Thai   ✓
    Note: In the following table, the checkmark indicates the languages each LLM model supports.