Configuration

curompt reads curompt.yaml (or the path passed via --config) to control providers, scoring thresholds, and report options.

Example

# ~/.config/curompt/curompt.yaml
storage:
  path: ~/.curompt/db.sqlite

providers:
  claude:
    api_key: $ANTHROPIC_API_KEY
    model: claude-3-7-sonnet
  openai:
    api_key: $OPENAI_API_KEY
    model: gpt-4.1-mini

analysis:
  max_tokens: 2000
  forbidden_terms:
    - "TBD"
    - "lorem ipsum"

evaluation:
  samples: 5
  schema: ./schemas/customer_onboarding.yaml
  temperature: 0.2

report:
  output: reports
  single_output: summary.md
  format: markdown

Storage

• storage.path: SQLite database storing run history, scores, and prompt snapshots. Defaults to ~/.curompt/db.sqlite.

Providers

Configure any provider you need under providers.<name>. The CLI maps these to subcommands, so curompt eval --provider claude will use the section above.

Analysis

• max_tokens: enforce prompt budget.
• forbidden_terms: list of phrases that should never appear (e.g., placeholders).

Evaluation

• samples: number of generations per prompt.
• schema: JSON Schema / OpenAPI file that responses must satisfy.
• temperature: provider-specific generation temperature.

Report

• output: directory for multi-file reports.
• single_output: optional all-in-one Markdown file.
• format: markdown or json.

CLI overrides

Every config value can be overridden via flags:

curompt scan --config team/curompt.yaml --output reports/ --single-output all.md
curompt eval --provider claude --samples 10

Use curompt config print (coming soon) to render the resolved configuration for debugging.