Cost & Token Usage

The AI processing consumes tokens (the unit of measurement for LLM usage), and costs vary with the volume of input material.
Factors affecting cost:
- Number and size of screenshots (each image-to-html conversion is a separate LLM call)
- Length of interview transcripts
- Size of the source codebase (larger codebases require more tokens for the application and database analysis)
- Model selection (more capable models cost more per token)
General guidance: a typical project with 20–40 screenshots, 2–3 interview transcripts, and a medium-sized codebase will cost in the range of tens of dollars for the full pipeline.
Tips for managing costs:
- Use the default model (claude-sonnet-4) for routine processing — it balances quality and cost well
- Process content incrementally during development and testing rather than running the full pipeline repeatedly
- Use the bash loop workaround for large file sets to maintain control over processing
- Review intermediate outputs before running the full analysis pipeline to avoid wasting tokens on poor inputs

All content is available under the Open Government Licence v3.0, except where otherwise stated

Support links