Tools that read your agent session logs - ccusage and alternatives

Your agents' logs are plain files, and a small ecosystem of open-source tools reads them. They answer different questions: what did it cost, what happened, what can I search.

#At a glance

All are MIT licensed except the groundedby evaluator, which runs in your browser on this site.

#ccusage

The best known of these. ccusage reads Claude Code's ~/.claude/projects (and ~/.config/claude/projects), and Codex's sessions and archived_sessions, compressed files included. It reports token usage and cost by day, week, month, session or five-hour billing block, and can feed a status line.

npx ccusage@latest daily
npx ccusage@latest codex daily

It moved from ryoppippi/ccusage to ccusage/ccusage, and the separate @ccusage/codex package is deprecated in favor of ccusage codex.

#claude-code-log

Turns Claude Code JSONL transcripts into readable HTML, Markdown or JSON, with a terminal interface and watch and serve modes. Good for archiving sessions in a form you can read years later.

#claude-code-viewer

A local web app for browsing Claude Code sessions, which can also start and resume them. Point it at another folder with --claude-dir.

#Agent Sessions

A macOS app that indexes sessions from many agents in one place, with search and resume. The widest format coverage of the group.

#splitrail and tokscale

Both track tokens and cost across many agents. splitrail can upload to its cloud if you turn that on. tokscale has an opt-in public leaderboard. Read their settings before running either.

#Before you run any of them

These tools read files that contain your code, keys and customers' data. Prefer tools you can read the source of, run them from a pinned version, and check whether anything is sent over the network. How to redact secrets from agent logs.