What the tool does
Three things: a public price board, a local quote of those prices, and regex redaction. It does not pick a model for you based on a measured quality score, and it does not send your prompt anywhere.
Price quotes
USD per 1,000,000 tokens, short-context list prices, refreshed on the 1st of each month.
llm-cost-opt quote --input-tokens 1500 --output-tokens 300
optimize can filter to rows whose editorial quality note is above a number you pass. That number is ours, not a benchmark.
Regex redaction
Emails, phone numbers, and IPv4. Optional long digit sequences. Names, postal addresses, and most free text are left alone.
llm-cost-opt anonymize --text "Email me at me@example.com"
Where it runs
CLI
JSON on stdout. Fine in a script.
Python
Import the quote and redaction helpers into your own process.
Local API
llm-cost-api serves /quote and /anonymize on localhost.