I've been tracking my AI CLI spending across Claude Code, Codex CLI, and Gemini CLI. Existing tools either don't support Codex at all or break on newer models like gpt-5.3-codex. So I built toktrack. One command npx toktrack What you get: - Per-model cost breakdown (gpt-5.3-codex, Opus, Sonnet, Gemini, etc.) - Daily/weekly/monthly trends - 52-week heatmap - Works across all three major AI CLIs Built in Rust with simd-json. Scans 3500+, 4GB+ session files in ~40ms. It also caches independently, so even after Claude Code deletes sessions (default 30 days), your cost history is preserved. GitHub: https://github.com/mag123c/toktrack MIT licensed. |
No comments yet