Configuration

GenDB is configured via gendb.yaml. Pass a custom path with --config.

Precedence

Settings are applied in this order (highest priority first):

1. gendb.yaml — file-based column instructions
2. LLM generation — the LLM generates data based on schema context and instructions

Full Reference

llm:
  provider: ollama        # ollama | openai | custom
  model: qwen2.5:7b       # Model name
  base_url: http://localhost:11434/v1  # LLM API endpoint
  api_key: ""             # API key (required for openai/custom)
  structured_output: true # Use JSON Schema to guarantee valid output
  temperature:            # Sampling temperature (optional, model default)
  chunk_size: 50          # Rows per LLM request

generation:
  default_rows: 100       # Rows per table unless overridden

  tables:                 # Per-table overrides
    users:
      rows: 500
      columns:
        bio:
          prompt: "Write a short professional bio"
        role:
          generator: one_of
          values: ["admin", "user", "moderator"]
    orders:
      rows: 2000

  column_rules:           # Pattern-based rules
    - pattern: "*_sku"
      generator: regex
      format: "[A-Z]{3}-[0-9]{6}"

Notes

1. llm.provider — LLM provider for data generation:

   • ollama (default) — local Ollama instance
   • openai — OpenAI API
   • custom — any OpenAI-compatible endpoint

2. llm.structured_output — When true (the default), GenDB sends a JSON Schema with LLM requests so the model is constrained to produce valid JSON. This dramatically reduces parsing failures, especially with local models. Set to false if your model or provider does not support structured output.

3. llm.chunk_size — Number of rows generated per LLM call. Larger values mean fewer API calls but longer responses. Default is 50.

4. generation.default_rows — Default number of rows to generate per table. Can be overridden per-table in the tables section.

5. generation.tables — Per-table configuration. Each table can specify a custom row count and per-column instructions for the LLM.

6. generation.column_rules — Pattern-based rules applied across all tables. Patterns use glob syntax (* for prefix/suffix matching). Rules are matched against column names.

Column Configuration

Each column override supports these fields:

Field	Description
generator	Override type: one_of, regex, or skip
prompt	Direct instruction to the LLM for this column
values	List of allowed values for one_of
format	Regex pattern for regex

Column Rules

Column rules apply instructions based on column name patterns across all tables:

column_rules:
  - pattern: "*_sku"
    generator: regex
    format: "[A-Z]{3}-[0-9]{6}"
  - pattern: "*_status"
    generator: skip

Patterns support * as a wildcard at the start, end, or both:

• *_email — matches columns ending with _email
• phone* — matches columns starting with phone
• *name* — matches columns containing name
• status — exact match only