Frugon

Frugon is a command line tool that analyzes what a team spends on language model calls. It reads request logs in JSONL format, in the style produced by the OpenAI and Anthropic APIs, counts the tokens, prices them, and reports which calls a cheaper model would have handled just as well and which should stay where they are. The analysis runs entirely on the local machine, makes no network calls in its default mode, and can emit an HTML savings report to hand to a team or a manager. It is aimed at engineering teams running production workloads across several models who suspect they are overpaying for routine traffic. Rather than inventing its own numbers, the tool assembles established components: token counting through tokencost, model prices through the LiteLLM registry refreshed weekly in continuous integration, and quality tiers anchored to public model arena rankings. A capture command collects logs for teams that have none, and a measure mode samples real prompts against the user's own providers using the user's own keys. It installs from PyPI with pip, pipx or uvx, runs on macOS, Linux and Windows, and supports Python 3.10 through 3.13. The analyzer is free and MIT licensed. A paid tier called Frugon Pro is published at $15 per month or $144 per year for an individual, and $39 per seat per month or $374 per seat per year for teams, with a three seat minimum. Pro adds a local daemon for continuous analysis after the command exits, a routing policy derived from the team's own logged traffic, and shared team reports. The site presents Pro through a waitlist with a rate locked for the life of the account, so general availability of the paid tier is worth confirming before planning around it.