Codex Token Calculator & Usage Check for ChatGPT Login

This repository provides a local Codex token calculator that allows you to perform an exact Codex token check even when you are signed in with a ChatGPT login instead of using OpenAI API keys. If you need a Codex CLI token calculator to figure out exactly how many tokens your projects are consuming locally, this tool reads your session logs and does the math for you.

Why this exists

If you use Codex through a ChatGPT login usage check flows are usually hidden because you don't have the standard API usage dashboard that comes with direct API billing. This repo fills that gap by acting as a read-only Codex token calculator. It parses the local session logs that Codex stores on disk and turns them into:

• An overall token total (Codex token check)
• A top-level per-project breakdown
• An exact per-directory breakdown
• A CSV export you can sort, filter, and graph later

Note: The script is entirely read-only. It calculates your Codex CLI usage locally and does not call the OpenAI API.

What the code does

codex_usage_report.py walks through ~/.codex/sessions, reads each .jsonl session log, and:

• Finds the working directory for the session from session_meta or turn_context
• Finds the last non-null token_count.total_token_usage snapshot
• Attributes that session to your chosen base directory
• Aggregates totals for:
  • Everything under the base directory
  • The first path segment under the base directory
  • The exact relative directory
• Writes all of it to codex_usage_report.csv

Older sessions are still counted in the summary if they are missing fields, but they are reported as missing attribution instead of being guessed.

Quick start

Clone or copy this repo somewhere convenient, then run the script against the directory tree you care about.

python3 codex_usage_report.py /path/to/workspace

If you are already inside the directory you want to summarize:

python3 /path/to/codex-usage-chatgpt-login/codex_usage_report.py .

Example run

Command:

python3 codex_usage_report.py /path/to/workspace --limit 5

Example terminal summary:

Base directory: /path/to/workspace
Sessions root:  /home/you/.codex/sessions

Summary
  session files seen:   312
  sessions in base:     284
  sessions attributed:  271
  missing usage:        13
  missing cwd:          7
  outside base:         21

These numbers are illustrative. Your real output will vary as your local session logs grow.

Example top-level table:

path	sessions	total_tokens	input_tokens	cached_input_tokens	output_tokens	reasoning_output_tokens
client-alpha	48	418,592,114	410,215,223	356,482,710	8,376,891	4,201,450
research-lab	22	191,042,887	187,998,552	154,201,984	3,044,335	1,622,948
web-redesign	35	129,774,502	127,420,111	102,840,320	2,354,391	1,211,674
automation	17	74,203,918	72,916,009	60,114,432	1,287,909	592,477
sandbox-demo	9	33,921,405	33,110,447	29,804,544	810,958	407,103

Example exact-directory table:

path	sessions	total_tokens
client-alpha/incident-review	8	84,772,440
research-lab/model-tuning	5	63,194,002
web-redesign/homepage-refresh	7	41,559,881
automation/reporting-pipeline	4	29,830,144
sandbox-demo/prototype-a	3	18,112,503

CSV output

Every run writes a CSV unless you pass --no-csv.

Default output file:

./codex_usage_report.csv

The CSV contains:

• one overall row with global totals and summary counters
• one row per top-level directory
• one row per exact relative directory

Example CSV columns:

section	path	sessions	total_tokens	input_tokens	cached_input_tokens	output_tokens	reasoning_output_tokens
overall	.	271	847,534,901	831,660,342	703,144,960	15,874,559	8,036,877
top_level	client-alpha	48	418,592,114	410,215,223	356,482,710	8,376,891	4,201,450
exact	client-alpha/incident-review	8	84,772,440	82,914,106	70,118,912	1,858,334	902,551

Useful options

Option	Purpose
--limit 20	Show only the first 20 rows in each printed table
--sort-by sessions	Sort by session count instead of total tokens
--csv-path report.csv	Write the CSV somewhere else
--no-csv	Skip CSV export
--json	Print the full report as JSON
--sessions-root /path/to/sessions	Use a different Codex session directory

Notes

• The report depends on what your local Codex client wrote to disk.
• Some older sessions may be missing cwd or token snapshots.
• Cached input tokens are reported separately because Codex logs them separately.
• This is a local estimate from session logs, not a billing statement from OpenAI.
• The generated CSV is intentionally gitignored because it can contain private folder names from your machine.