Configuration

Learn how to write config files for classify.

Config Structure

settings:
  model: claude-sonnet-4-5-20250929
  reasoning: true
  batch_size: 10000

input:
  file: data.csv
  columns: [title, description]
  id_column: optional_id_column

prompt:
  system: "You are an expert at categorizing content."
  template: |
    Categorize this:

    Title: {title}
    Description: {description}
output:
  fields:
    - name: category
      type: string
      description: "Content category"
      enum: ["tech", "sports", "other"]

Settings

model

Required

The Claude model to use. Current options:

• claude-sonnet-4-5-20250929 - Best balance of speed and quality
• claude-haiku-4-5-20250929 - Fastest, cheapest

reasoning

Optional - Default: false

When enabled, adds {field}_reasoning columns to output explaining each classification.

settings:
  reasoning: true  # Adds category_reasoning, confidence_reasoning, etc.

batch_size

Optional - Default: 10000

Maximum number of requests per batch. Claude Batch API has a 100,000 request limit.

settings:
  batch_size: 10000  # Good for most use cases

Input

file

Required

Path to your CSV file.

columns

Required

List of column names to include in the prompt. Only these columns are sent to the API.

input:
  file: data.csv
  columns: [title, description, author]  # Only these 3 columns

id_column

Optional

Use an existing column as the unique identifier instead of generating one.

input:
  file: data.csv
  columns: [title, description]
  id_column: user_id  # Must be unique

Prompt

system

Required

System prompt describing the task. Be specific about what you want classified.

prompt:
  system: "You are an expert at categorizing content. Classify each item into one of the allowed categories."

template

Required

Template for formatting each row. Use {column_name} placeholders.

prompt:
  template: |
    Categorize this content:

    Title: {title}
    Description: {description}
    Author: {author}

Output

fields

Required - List of output fields to generate.

Each field needs:

Property	Required	Description
name	Yes	Field name (lowercase, no spaces)
type	Yes	string, integer, number, boolean
description	Yes	What the field represents
enum	No	Allowed values (for strings only)

output:
  fields:
    - name: sentiment
      type: string
      description: "Overall sentiment of the content"
      enum: ["positive", "negative", "neutral"]

    - name: score
      type: integer
      description: "Confidence score from 1-10"

    - name: urgency
      type: number
      description: "Urgency level from 0.0 to 1.0"

    - name: is_flagged
      type: boolean
      description: "Whether content should be flagged for review"

Complete Example

settings:
  model: claude-sonnet-4-5-20250929
  reasoning: true
  batch_size: 10000

input:
  file: reviews.csv
  columns: [product_name, review_text, rating]

prompt:
  system: "You are a sentiment analysis expert. Analyze product reviews and classify sentiment, extract key themes, and assess review quality."

  template: |
    Analyze this product review:

    Product: {product_name}
    Rating: {rating}/5
    Review: {review_text}

    Provide sentiment analysis and key themes.

output:
  fields:
    - name: sentiment
      type: string
      description: "Overall sentiment: positive, negative, or neutral"
      enum: ["positive", "negative", "neutral"]

    - name: themes
      type: string
      description: "Comma-separated list of key themes mentioned"

    - name: quality_score
      type: integer
      description: "Review quality score from 1-5 based on detail and usefulness"

Validation

Always validate your config before running:

classify check my_config.yaml

This will: - Check CSV file exists and is readable - Verify all referenced columns exist - Validate output schema - Calculate estimated costs - Show sample rows that will be processed