Cherry Bridge AI

Free open-source resource · Windows preview

Track AI usage against the outcome you actually want

Cherry Meter brings Claude Code and Codex usage into one local report per device. See tokens by provider and model, keep your own outcome checkpoints, and check what the evidence actually covers.

Usage, targets and outcomes

Usage tells you what was consumed. You define what progress means.

The tool keeps measurements and human judgment visibly separate, so a busy week never gets mistaken for a useful result.

Observed usage

See observed Claude Code and Codex tokens together, separated by provider and model, with source age, attribution and coverage details.

Advisory target

Keep a token target you choose alongside observed usage. Targets are advisory and never pause or gate work; comparisons depend on qualified coverage.

Outcome checkpoints

Record progress only when you say it happened. Usage and assistant activity never advance an outcome for you.

A useful first view

Open a report you can inspect and keep

Reports are saved as private local HTML and machine-readable evidence. Refreshing creates a new snapshot, while the earlier view stays readable.

  • Dated snapshots, source age, model breakdown and coverage labels
  • Explicit partial coverage and unassigned attribution where needed
  • Explicit checkpoint history with operator attribution

Synthetic example

My device report

Complete coverage
Seven-day usage

62,400

of 100,000 advisory tokens

Outcome progress

Prepare the first user-ready release

1 of 2 checkpoints complete

Operator-reported

Get started with 1.2.1-rc.1

Install the preview, then approve your measurement sources

Windows AMD64 preview

Add the Claude Code plugin

Use existing trusted Python 3.10+, Git and Claude Code CLI. The installation guide adds our self-hosted marketplace and sets an explicit, verified Python path before the first session. Check for an existing installation first so you do not enable two copies.

Follow the installation guide

Connect your local sources

Codex measurement reads only approved local session roots. Prospective Claude Code metrics need a separately configured local collector and provider export settings. Installing the plugin alone does not enable collection or install that helper.

Read the device setup steps

Local measurement, explicit sharing

You approve the source paths and report writes. The measurement engine makes no remote requests and does not retain prompt or response text. When you use the plugin through Claude, selected tool output and reports may enter your hosted conversation under your Anthropic settings. Keep private reports and configuration out of public issues.

What is verified today

  • Controlled native Windows Claude Code and Codex measurement, with 1.2.0 client evidence
  • Windows preview install, version update and rollback checks
  • Clearly labelled synthetic demos and dated real-usage evidence

This is a preview, not a stable cross-platform release. Native Mac/ARM, complete daily capture, sleep/reboot continuity and all IDE/GUI paths remain unqualified. Reports can have partial coverage and unassigned Claude attribution. Tokens are not billed cash or proof that an outcome was achieved.

Use it, update it, know its limits

Five commands in Claude Code

Setup, demo, report, checkpoint and doctor use the plugin's declared local MCP server. Actions that read or write your data show a plan for approval. Start with a synthetic demo; its 300 tokens demonstrate the report, not your actual usage. A saved report is a snapshot, so generate a new one to see later activity.

Read the command and data-access contract

Deliberate updates and rollback

Review each new version and its checksum before updating. This preview does not enable auto-update, and its version-specific marketplace branch will not automatically follow a future release. Keep the previous source, interpreter settings and private data for rollback.

Read the update and rollback steps

Independent distribution

You can download this public preview from GitHub or add its project-owned marketplace. It is not an approved Anthropic Directory listing. The separate 1.2.0 submission is held by the directory's Verified-organization eligibility rules for local MCP servers and code files.

Read the directory status

Keep the evidence in view

Controlled live-provider results and synthetic packaging tests answer different questions. Hosted Windows Server checks do not establish a clean consumer, non-admin installation. Ordinary Claude chat and Cowork are outside this measurement scope.

Read the qualification record

Already using the original WSL/Codex tool? Stable 1.0.1 remains available. Keep its archive and setup prompt together. The preview is a separate install; it does not silently migrate your existing setup.

Review data handling or browse troubleshooting and support before sharing diagnostics.

Ready to compare activity with progress?

Start with the versioned Windows preview and a synthetic demo. Then connect only the real sources you approve and inspect a new report snapshot when you want updated usage.

Download Windows preview