Agent Usage Stat

Agent Usage Stat

A private desktop analytics app for understanding how Claude Code, OpenAI Codex, GitHub Copilot CLI, and opencode use tokens, time, and API-equivalent cost.

Agent Usage Stat turns local coding-agent transcripts into one searchable usage ledger. Compare providers and models, follow spend and token volume, inspect projects and sessions, and see when work happened. Prompt and response text never enters the ledger.


Supported

• Windows and macOS
• Claude Code
• OpenAI Codex, including Codex sessions used through ChatGPT
• GitHub Copilot CLI
• opencode

Agent Usage Stat does not read general ChatGPT or Claude.ai chats. All usage data stays on the local machine or in the folder selected by the user.

Install

Download the current installer from [GitHub Releases](https://github.com/yiliangs/agent-usage-stat/releases):

• Windows installer: Agent-Usage-Stat-Setup.exe
• Windows portable: Agent.Usage.Stat-win32-x64-*.zip
• macOS, Apple Silicon: Agent.Usage.Stat-*-arm64.dmg
• macOS portable, Apple Silicon: Agent.Usage.Stat-darwin-arm64-*.zip

The macOS build is Apple Silicon only. Intel Macs are not covered by a release yet, because the bundled capture helper is a single executable built for one architecture at a time.

No separate Node.js or npm installation is required.

The macOS build is ad-hoc signed rather than notarized, so Gatekeeper blocks the first launch. After dragging the application into `/Applications`, open it once with Control-click, then **Open**. If macOS still refuses, clear the download quarantine flag and try again:

xattr -dr com.apple.quarantine "/Applications/Agent Usage Stat.app"

On first launch, choose where to keep the durable usage ledger and how sessions should be captured. Agent Usage Stat then reconciles existing sess