codex-usage-analyzer

codex-usage-analyzer

Read the account usage shown by Codex through the official app-server protocol, from one small CLI.

codex-usage-analyzer starts a compatible Codex process from your installed CLI or macOS app, calls account/usage/read, and emits a stable, identity-free contract. It does not scan local sessions or directly read authentication files, tokens, keychains, prompts, or responses.

Documented upstream: This CLI uses OpenAI Codex's documented account/usage/read app-server method.

Support

Maintained with support from OpenAI’s Codex for Open Source program.

Support is provided to the maintainer and does not imply endorsement.

Quick start

npx codex-usage-analyzer@latest

Human-readable output:

Codex account usage

Lifetime tokens    1.23B
Peak daily tokens  45.6M
Longest turn       12m 34s
Current streak     3 days
Longest streak     21 days
Daily buckets      30 days

Captured at 2026-07-11T00:00:00.000Z

The values above are synthetic. Your command reads the usage available to the currently signed-in Codex account.

For machine-readable output:

npx codex-usage-analyzer@latest --json

{
  "contractVersion": 1,
  "capturedAt": "2026-07-11T00:00:00.000Z",
  "summary": {
    "lifetimeTokens": 1234567890,
    "peakDailyTokens": 45600000,
    "longestRunningTurnSec": 754,
    "currentStreakDays": 3,
    "longestStreakDays": 21
  },
  "dailyUsageBuckets": [
    {
      "startDate": "2026-07-10",
      "tokens": 123456
    }
  ]
}

Why this CLI

• Account-level source: use the same app-server method intended for Codex account usage instead of estimating from retained local files.
• Privacy-first boundary: receive usage metrics without adding names, usernames, avatars, emails, account identifiers, or credentials to the output.
• Stable integration: consume a versioned JSON contract with allowlisted fields and explicit null semantics.
• Small runtime: use Node.js built-ins and a Codex CLI or compatible macOS app already on your machine; there are no runtime package dependencies.

Codex lookup benchmark

Lower is faster. This snapshot measured one warm environment on 2026-07-13 with one warm-up and five timed runs.

Tool	Median	Relative time
codex-usage-analyzer@0.2.0	1.145s	1.0x
ccusage@20.0.17	3.306s	2.9x
tokscale@4.4.1	19.723s	17.2x

This compares Codex usage lookup only, not whole products or every workflow they support. The data sources also differ: codex-usage-analyzer reads account-level usage through app-server, while the comparison commands scan retained local Codex session history. See the benchmark methodology, environment, command arrays, and limitations.

Supported metrics

Field	Meaning
summary.lifetimeTokens	Lifetime token usage
summary.peakDailyTokens	Highest token usage reported for one day
summary.longestRunningTurnSec	Longest-running turn, in seconds
summary.currentStreakDays	Current activity streak, in days
summary.longestStreakDays	Longest activity streak, in days
dailyUsageBuckets	Source-dated daily token buckets, when available

Every summary field is present. A value of null means the upstream method did not provide that metric; it does not mean zero. Daily buckets can likewise be null, an empty array, or an array of dated values.

Requirements

• Node.js 20 or newer
• A recent Codex CLI available as codex on PATH, or a compatible macOS ChatGPT app or Codex app installed in a standard Applications location
• A ChatGPT-backed Codex sign-in that supports account/usage/read

API-key-only and Bedrock authentication do not provide this account usage method. Sign in through Codex before running the analyzer. The package delegates authentication to the selected Codex CLI or app and never asks you to paste a token.

CLI reference

codex-usage-analyzer - Read your Codex account usage

Usage:
  codex-usage-analyzer [usage] [--json]
  codex-usage-analyzer [usage] --help
  codex-usage-analyzer --version

Command	Output
codex-usage-analyzer	Human-readable account usage
codex-usage-analyzer usage	Same human-readable output
codex-usage-analyzer --json	Account Usage Contract JSON
codex-usage-analyzer usage --json	Same JSON output
codex-usage-analyzer --help	Help without starting app-server
codex-usage-analyzer --version	Package version without starting app-server

Successful output is written to stdout. Failures are written to stderr as a stable error code and a safe message, without raw RPC data or app-server stderr.

SDK

import {
  ACCOUNT_USAGE_CONTRACT_VERSION,
  CodexUsageError,
  readAccountUsage
} from "codex-usage-analyzer";

try {
  const usage = await readAccountUsage({ timeoutMs: 15_000 });
  console.log(usage.contractVersion === ACCOUNT_USAGE_CONTRACT_VERSION);
} catch (error) {
  if (error instanceof CodexUsageError) {
    console.error(error.code);
  }
}

The SDK returns the same document as CLI --json. See the Account Usage Contract and JSON Schema for field and compatibility rules.

How it works

• Prefer the codex executable available on PATH.
• On macOS, fall back to a compatible ChatGPT or Codex app bundle in a standard Applications location.
• Spawn the selected Codex executable with app-server, without a shell.
• Complete the stable app-server initialization handshake.
• Call account/usage/read.
• Allowlist and validate the supported fields.
• Stop the child process and return the normalized document.

The package has no direct credential reader and no private profile endpoint fallback. Authentication and service communication remain inside the selected Codex process.

Downstream integrations

Profile sites, README cards, and other services can accept the identity-free JSON contract and combine it with identity they manage separately. The Downstream Integration Guide defines recommended field names, ownership, submit-token, validation, rendering, caching, and deletion boundaries.

Do not add identity fields to the account usage document. A downstream service should resolve GitHub identity from its own authenticated account binding, not trust a display name or avatar submitted by this CLI.

Privacy and Security

The default CLI does not directly read or emit:

• access tokens, refresh tokens, cookies, or keychain entries
• names, usernames, avatars, emails, or account identifiers
• prompts, responses, tool input, tool output, or local session files
• local filesystem paths, raw RPC responses, or raw app-server stderr

Treat account usage as private data even though the contract excludes identity. Review a downstream service's retention and visibility policy before submitting output anywhere.

For vulnerability reporting and supported versions, see SECURITY.md.

Troubleshooting

Error code	What to check
CODEX_NOT_FOUND	Install or update a Codex CLI on PATH, or install a compatible macOS ChatGPT or Codex app in a standard Applications location.
APP_SERVER_START_FAILED or APP_SERVER_EXITED	Confirm the selected Codex CLI or app can start and that your environment permits child processes.
APP_SERVER_TIMEOUT	Retry after checking connectivity; SDK callers can set timeoutMs up to 120000.
APP_SERVER_RPC_ERROR	Update Codex and confirm a compatible ChatGPT-backed sign-in.
APP_SERVER_PROTOCOL_ERROR or INVALID_ACCOUNT_USAGE_RESPONSE	Update both Codex and this package; the upstream response was not safe to normalize.

The CLI intentionally suppresses raw upstream details. When reporting a bug, include the package version, Codex version, Node.js version, platform, command shape, and error code only.

Development

npm test
npm pack --dry-run

Maintainers should use the repository's npm release guide for release operations; those steps are intentionally kept out of the user guide.

Contributing

Read CONTRIBUTING.md before opening a pull request. Changes to the public contract need an issue and explicit consumer-impact review.

Support

Use GitHub Issues for reproducible bugs and focused feature requests. Use GitHub's private vulnerability reporting path for security-sensitive reports.

License

Copyright (c) postmelee. Released under the MIT License.

This license covers only this repository's code and documentation. It does not grant rights to OpenAI services, Codex assets, user data, model outputs, trademarks, or third-party content.

This independent project is not affiliated with, endorsed by, or sponsored by OpenAI. OpenAI and Codex names and trademarks belong to their respective owners.