cadence.report

What It Is

Cadence is a local time-utilization reporting tool for agentic coding. Its core use case is helping developers understand where their Claude Code and Codex “agent hours” are going, how much they cost, and what they produced. It scans existing JSONL logs on your machine, writes a SQLite cache under ~/.cache/cadence, and generates static HTML reports with views for time, cost, output, efficiency, project breakdowns, and more.

Core Capabilities and Technical Boundaries

Its key differentiator is an offline-first design: report generation makes zero network calls, requires no account linking, and includes no telemetry, remote fonts, or remote charts. It also states that it does not read source code files, only usage logs, with optional local git diff stats. Its metrics go beyond active hours, lines of code, and merged PRs, emphasizing measures that better reflect AI coding-agent utilization, such as cost per shipped hour, active-vs-idle ratio, and cache efficiency. It requires Python 3.11+, plans installation via a macOS Brew tap or a shiv .pyz zipapp, and provides --strict to turn parsing warnings into hard errors in CI.

Pricing and Team Edition

The personal Solo edition is listed as a free, open-source CLI for single-machine, single-account use. It requires no account and includes full local reports plus a month-end Cadence Wrapped. The Team edition is $12/seat/month and aggregates multiple machines and accounts into a central dashboard, with filtering by device and account as well as side-by-side comparisons. However, the price is marked as indicative, and the product is still in a launching soon / waitlist stage.

Pros and Cons

The advantages are clear privacy boundaries and suitability for sensitive code environments, while offering a stronger focus than WakaTime or GitHub activity stats on the relationship between AI-agent cost, efficiency, and output. The downsides are also obvious: it is not yet available, and the ecosystem currently only explicitly covers Claude Code and Codex. Beyond macOS, Python 3.11+, and Brew/.pyz, the cross-platform experience is unknown. API/SDK availability, self-hosted team dashboards, payment methods, and formal support policies have not been disclosed.

Who It’s For and Access from China

Cadence is suitable for independent developers who heavily use Claude Code/Codex, early-stage teams building around AI coding tools, and organizations that want to keep prompts and source code local while only aggregating statistical data. Its accessibility from China cannot be determined from the available text. If actual usage depends on Claude/Codex, it may be affected by upstream network and payment restrictions. Alternatives include WakaTime and GitHub activity/insights, though they do not fully cover the agent utilization and cost-output retrospectives that Cadence emphasizes.