# Tokenmin

> Open-source Claude usage advisor. Reads every Anthropic optimization doc (caching, parallel tools, MCP hygiene, model routing), watches your actual Claude usage, and shows you the next dollar you can save. Friends-and-family preview is free in exchange for anonymized usage data; the scanner that does the anonymization is Apache-2.0 and publicly auditable.

## Install

- [install.sh](https://tokenmin.ai/install.sh): one-line installer. `curl --proto '=https' --tlsv1.2 -fsSL https://tokenmin.ai/install.sh | bash`
- [install.sh.sha256](https://tokenmin.ai/install.sh.sha256): SHA-256 of the installer for verify-then-run.

## Source

- [tokenmin-scanner](https://github.com/watsonrm/tokenmin-scanner): the public, Apache-2.0 audit copy. Collector + anonymizer + CLI. About 5 minutes to read end-to-end.
- [tokenmin-site](https://github.com/watsonrm/tokenmin-site): this site.

## Trust

- [SECURITY.md](https://github.com/watsonrm/tokenmin-scanner/blob/main/SECURITY.md): threat model + disclosure path. security@rmwcommerce.com, 14-day SLA for high-severity.
- [Anonymization design](https://github.com/watsonrm/tokenmin-scanner#anonymization): HMAC-SHA256 with per-install salt, 16-char hash output, 64 KiB input cap.

## Watched sources

- [sources.json](https://github.com/watsonrm/tokenmin-scanner/blob/main/bin/sources.json): the curated list of trusted Claude-optimization sources tokenmin's detector-research watcher monitors weekly. Tier 1 (Anthropic first-party) and Tier 2 (verified community) with trust_reason on every entry.