Hi all - I built tokenusage (tu), a blazing-fast local usage tracker with unified Codex + Claude reporting.

What it focuses on:

• Speed on large local logs
• Merged totals/model/cost across providers
• Multiple views: CLI / TUI / GUI

Implementation summary:

• Rust-native pipeline
• parallel file discovery and parsing
• incremental cache reuse
• source adapters for Codex/Claude logs
• merged aggregation for day/week/month outputs

Benchmark method (real local data, Codex-only):

• Machine: Apple M3 Max, macOS 15.6.1
• Dataset: ~/.codex/sessions (71 JSONL files, ~537 MB)
• Date range: 2025-09-01 to 2026-02-28
• tu 1.1.2 vs @ccusage/codex 18.0.8

Results:

• Cold: tu 0.19s vs ccusage-codex 6.56s (~34.5x)
• Warm (avg of 10 runs): tu 0.052s vs ccusage-codex 6.819s (~131.1x)

Install:

cargo install tokenusage --bin tu
npm install -g tokenusage

Links:

• crates package
• npm package
• Screenshots in the GitHub README

Feedback welcome, especially on parsing accuracy and cache strategy.

1 post - 1 participant

Read full topic