AI TechSeptember 10, 2026 · 8 min

ccusage: The Agent Bill, Counted from Local Logs

This topic explains how to cut spending but never showed how to count it. This command reads the logs of sixteen agent CLIs straight from disk and turns them into reports by day, session and five-hour window - with no install and no data leaving your machine.

ccusage - расход и стоимость работы агентских CLI, собранные из локальных логов

This topic already has articles on cutting context and not overpaying. One thing was missing: how to see what you are actually spending. Optimising blind is a dubious pastime.

ccusage closes exactly that gap. It reads the logs that agent command-line tools already write to disk and turns them into reports by day, week, month and session. It sends nothing anywhere: the data is already yours, and the tool only counts it.

It was written by @ryoppippi, ships under MIT, and as of 25 August 2026 has 18,152 stars and version 20.0.20. The project is alive: the last commit landed the same day I wrote this.

It Runs Without Installing

This comes first because it removes half the questions. There is no need to install globally, and the documentation says so.

bunx is recommended: it caches the downloaded package, so repeated runs are faster than the first. If you do want it globally, the usual npm, bun, yarn and pnpm routes exist.

A pleasant detail of its own - builds from not-yet-merged changes, so you can test a fix before the release:

Sixteen Sources

It reads more than Claude Code. The list as of 25 August 2026: Claude Code, Codex, OpenCode, Amp, Droid, Codebuff, Hermes Agent, pi-agent, Goose, OpenClaw, Kilo, Kimi, Qwen, GitHub Copilot CLI, Gemini CLI and Grok Build CLI. Sixteen.

That works in two modes. The unified report gathers everything found on the machine into one table. The named one narrows the same report to a single source and unlocks its specific options.

Where the logs live is documented too. For Codex, for instance, it is the directory in the variable - by default, but you can list several separated by commas, and inside each it finds both active and archived sessions on its own.

AI TechSeptember 10, 2026 · 8 min

ccusage: The Agent Bill, Counted from Local Logs

This topic already has articles on cutting context and not overpaying. One thing was missing: how to see what you are actually spending. Optimising blind is a dubious pastime.

It was written by @ryoppippi, ships under MIT, and as of 25 August 2026 has 18,152 stars and version 20.0.20. The project is alive: the last commit landed the same day I wrote this.

It Runs Without Installing

This comes first because it removes half the questions. There is no need to install globally, and the documentation says so.

bunx is recommended: it caches the downloaded package, so repeated runs are faster than the first. If you do want it globally, the usual npm, bun, yarn and pnpm routes exist.

A pleasant detail of its own - builds from not-yet-merged changes, so you can test a fix before the release:

Sixteen Sources

That works in two modes. The unified report gathers everything found on the machine into one table. The named one narrows the same report to a single source and unlocks its specific options.

Mode	What it does
`auto`	The default: where a ready cost exists it uses it, where it does not it calculates from tokens
`calculate`	Always calculates from tokens, ignoring the ready values
`display`	Shows only what has already been calculated, adding nothing

ccusage: The Agent Bill, Counted from Local Logs

It Runs Without Installing

Sixteen Sources

Sections

More

Connect

ccusage: The Agent Bill, Counted from Local Logs

It Runs Without Installing

Sixteen Sources

The Reports

Five-Hour Windows

Three Cost Calculation Modes

Codex Tiers and Honesty in a Small Thing

Filters and Output

The Data Goes Nowhere

What to Know in Advance

First-Encounter Mistakes

Sources