Bulk Processing Operations

Hubrix supports five operation types for bulk processing. Each operation processes one CSV row at a time and writes one or more output columns to the results.

Step 1 of the bulk job wizard showing the five operation type cards
Step 1 of the bulk job wizard showing the five operation type cards

Prompt

Run a custom AI prompt for every row, using column values as variables.

Config options:

  • Model — choose which AI model to use (GPT-4o, Claude, Gemini, etc.)
  • Prompt template — write your prompt using {{column_name}} placeholders, e.g. Write a 50-word product description for {{product_name}} priced at {{price}}.
  • Output column name — the name of the column added to the results

Best for: Writing product descriptions, summarising support tickets, generating personalised outreach copy, scoring or tagging free-text fields.

The Prompt operation is the most flexible. Any task that can be expressed as a fill-in-the-blank prompt is a good candidate.


Agent

Route each row through one of your Custom Agents.

Config options:

  • Agent — select which agent to use from your library
  • Input column — the CSV column sent to the agent as the user message

Output: The agent's full reply is added as a new column.

Best for: Rows where you want the agent's system prompt, tools (web search, code execution), and knowledge base to all be applied to each row's data.


Classify

Assign each row to one of your defined categories.

Config options:

  • Text column — the column containing the text to classify
  • Categories — define a list of category labels (e.g. Billing, Technical Issue, Feature Request, Other)
  • Category descriptions (optional) — a brief description of each category helps the AI classify more accurately

Output: A category column is added with the assigned label.

Best for: Support ticket triage, lead scoring bins, content tagging, sentiment buckets (Positive / Neutral / Negative).


Extract

Pull structured fields from each row's text column.

Config options:

  • Text column — the column containing the unstructured text to extract from
  • Fields to extract — define each field name and a description (e.g. vendor_name: The name of the company that issued the invoice)

Output: One new column is added per extracted field.

Best for: Extracting structured data from invoices, contracts, emails, job descriptions, or any unstructured text.


Translate

Translate a text column into a target language.

Config options:

  • Text column — the column to translate
  • Target language — select from a list of supported languages (e.g. English, French, Spanish, German, Japanese)
  • Preserve formatting — if enabled, HTML or Markdown formatting in the source text is preserved in the translation

Output: A translation column is added with the translated text.

Best for: Translating product descriptions, customer reviews, support tickets, or survey responses.

Translation quality depends on the underlying model and the source language. For major languages (English, French, Spanish, German, etc.) quality is high. For lower-resource languages, review a sample of results before using them in production.

Was this helpful?