Which prompt cost the most?
TokenAtlas shows what each prompt cost across Claude Code, Codex, OpenCode and Pi, and keeps the history on your machine even after the agents clean up their own logs. Below is a real report built from fictional logs. It runs entirely in your browser, with no network requests.
Open the demo report- Start with Analytics: compare token usage over time and change the breakdown.
- Open Show context on a turn for its branch, title and final message.
- See the share of the 5-hour and weekly limit on each turn, and Limit hits further down.
- Filter by agent, project or model at the top.
- Open Token efficiency for calculated facts and their evidence. Scenarios show what a selected assumption would imply, not proven savings.
Open the report full screen About TokenAtlas
Try it with your own logs: pipx install tokenatlas then tokenatlas open.
Open source (MIT) on GitHub