See what used up your Claude Code or Codex limit.

OpenReason shows where your Claude Code and Codex allowance went, turn by turn. Free for you. One view of cost and output for your team.

Open Session Forensics Start free

Forensics reads your files in your browser. No sign-up. Nothing is uploaded.

Claude Code
Codex
Mac
Windows
Linux
Not a proxy

Your files stay on your computer.
Your report is created here.
Nothing is uploaded.

Choose the coding tool you want to analyze

Claude Code~/.claude/projectsChoose folder

Codex~/.codex/sessionsChoose folder

Codex · 15 files · 1,240 requests · 212 readings

Limits reached2· last 7 days

Pick one to see what used it up.

Nov 9, 2026, 15:07Reset 16:03Codex · 5-hour100% recordedBlocked
Nov 5, 2026, 18:41Reset 20:12Codex · 5-hour100% recordedBlocked

What used it up3 turns

3 requests in one chat used about 31% of your 5-hour allowance.

Every step re-reads the whole chat, so a long chat makes every step expensive.

14:43+14% of 5-hour
14:52+10% of 5-hour
15:00+7% of 5-hour

Example data

Example team

OwnerMIMaya Iyer

SpendNovember 2026

Committed$1,15014 seats on 2 plans

Variable~$258.40Claude overage · Codex credits

Total~$1,408November 2026

Line	Committed	Variable	Total
Anthropic
Claude Team8 Premium × $125 · monthly	$1,000	—	$1,000
Claude overage3 people went past quota		$212.40	$212.40
OpenAI
ChatGPT Business6 Standard × $25 · monthly	$150.00	—	$150.00
Codex credits1150 credits × $0.04 · 1 person went past quota		~$46.00	~$46.00
November 2026	$1,150	~$258.40	~$1,408

Claude Team$1,000
8 Premium × $125 · monthly
Claude overage$212.40
3 people went past quota
ChatGPT Business$150.00
6 Standard × $25 · monthly
Codex credits~$46.00
1150 credits × $0.04 · 1 person went past quota
November 2026~$1,408

Committed: seats seen this month at your plan prices. Variable: the vendor's own figure where it reports one; ~ marks our estimate at list price.

Example team · November 2026

Session Forensics · free

Find the turns that ate your 5-hour window.

Point it at your Claude Code or Codex folder. It reads the files in your browser and shows which limits you hit, which turns took the most, and when you work.

The Claude Code 5-hour figure is an estimate. For Codex you need a paid plan, which records a 5-hour window.

What to do when you hit your limit All guides

From Codex's own 5-hour meter.

Intel, on your own

See how close you are, from every machine you use.

Install or-otel and every turn is kept, with its estimated share of your 5-hour and weekly windows and what it cost once the window was full.

For engineers Docs (opens in a new tab)

Example team

OwnerMIMaya Iyer

Closest to quota: 84%, Anthropic · 5-hour

Then bring your team.

The first person from your company creates the workspace. Teammates on the same email domain ask to join, or you invite them. The lead then sees every seat, every overage, and the work it went into, across both agents, without telling anyone which one to use.

Example team

OwnerMIMaya Iyer

Overage incidents

Who ran past their plan, and why.

Every time someone went past a limit and kept working: who, which session, the likely cause, and the commit or pull request it produced, if any. Open one to see every turn.

Example team

OwnerMIMaya Iyer

Overage incidentsNovember 2026

Incidents4November 2026 · 4 people

Overage~$258.40Some incidents estimated at list price

Produced something3 of 4A commit, PR or merge inside the interval itself

Time	Person	Session · repo	Cause	Produced	Cost
Nov 4, 14:1338 min before reset	Priya Nandakumar	a3f9c1d2acme/api-gateway · 61 requests	Long context	Commit a41f09c2 in acme/api-gateway	$96.20
Nov 6, 10:4852 min before reset	Tom Okafor	7be20c95acme/web-app · 88 requests	Parallel burst	Pull request in acme/web-app	$71.80
Nov 7, 16:221 h 10 min before reset	Lena Brandt	c41d7e08acme/worker · 34 requests	Long context	Commit 7c2e51d0 in acme/worker	~$46.00
Nov 9, 09:0524 min before reset	Dev Raman	e90b3a6facme/ledger · 29 requests	Cold restart	Nothing detected	$44.40

Priya Nandakumar$96.20Nov 4, 14:13 · 38 min before resetacme/api-gateway · 61 requestsLong contextProduced: Commit a41f09c2 in acme/api-gateway
Tom Okafor$71.80Nov 6, 10:48 · 52 min before resetacme/web-app · 88 requestsParallel burstProduced: Pull request in acme/web-app
Lena Brandt~$46.00Nov 7, 16:22 · 1 h 10 min before resetacme/worker · 34 requestsLong contextProduced: Commit 7c2e51d0 in acme/worker
Dev Raman$44.40Nov 9, 09:05 · 24 min before resetacme/ledger · 29 requestsCold restartProduced: Nothing detected

Cost: the vendor's own figure where it reports one; ~ marks our estimate at list price. Cause: hover a tag for the rule behind it. Produced: a commit, PR or merge recorded inside the interval — work the session did before or after it does not count.

Spend · November 2026 › Overage incidents › Priya Nandakumar · Tue 4 Nov

Priya Nandakumar · Tue 4 Nov

Overage on Claude Code from 14:13 to 14:51.

Overage$96.20the vendor's own figure

Requests past quota61104 in the window

Before reset38 minWindow reset at 14:51

Quota

The 5-hour window: how full the quota got, and what it cost once it was full.

5-hour quota

Overage

Quota exhausted 14:13Window reset 14:51

Turns3 turns · 61 requests

Time	What was asked	Context	Cost
14:16	retry the failed invoice webhooks and log each attempt	412k	$38.40
14:29	now make the retries back off and stop after five	438k	$31.10
14:44	add a test for the give-up case and run the suite	455k	$26.70

Example team · November 2026

Cause tags are patterns to investigate, not proof of waste. Only work recorded inside the overage window counts as produced.

Work

From quota to pull request, feature by feature.

Each branch with the agent time, quota and overage it took, and the pull request it went into, open or merged.

Example team

OwnerMIMaya Iyer

Work

Person	Feature	Agent time	Turns	Quota	Overage	Shipped
Ppriya	feat/invoice-retries	~3:10:00	64	38%	$96.20	#212
Llena	feat/export-queue	2:40:00	52	29%	~$46.00	#209
Ttom	fix/session-timeout	1:25:00	31	17%	$71.80	PR open
Ssofia	main (no branch)	0:35:00	12	6%	$0	—

feat/invoice-retries~3:10:00agent timePpriya64 turns · Quota 38% · Overage $96.20#212
feat/export-queue2:40:00agent timeLlena52 turns · Quota 29% · Overage ~$46.00#209
fix/session-timeout1:25:00agent timeTtom31 turns · Quota 17% · Overage $71.80PR open
main (no branch)0:35:00agent timeSsofia12 turns · Quota 6% · Overage $0—

Example team · November 2026

Agent time is the agent's working time, not the engineer's. ~ marks a reconstructed or estimated figure.

One command per machine.

Your requests still go straight to Anthropic and OpenAI, on the plans you already have. OpenReason is not a proxy.

Run it once

It installs or-otel and sets up Claude Code and Codex to report their usage. Mac, Windows or Linux, WSL included.

$ curl -fsSL https://get.openreason.app/otel \    | OR_API_KEY=or_ingest_... bash  ✓ Setup complete.

Paste it into Terminal. It asks for your Mac password once.

PS> $env:OR_API_KEY = 'or_ingest_...';    irm https://get.openreason.app/otel.ps1 | iex  ✓ Setup complete.

PowerShell. Windows asks once whether to allow the change.

C:\> set "OR_API_KEY=or_ingest_..." && curl -fsSL ^       https://get.openreason.app/otel.cmd -o %TEMP%\or-otel.cmd ^       && %TEMP%\or-otel.cmd && del %TEMP%\or-otel.cmd  ✓ Setup complete.

Command Prompt, the same install.

$ curl -fsSL https://get.openreason.app/otel-linux \    | OR_API_KEY=or_ingest_... bash  ✓ Setup complete.

Any terminal, WSL included, where it installs inside WSL. It asks for your password once.

Send one prompt
Each agent shows Connected. Then invite your team; each teammate runs their own command.

One command removes all of it.

On Claude Team or Enterprise? Connect Claude Code for everyone with one organisation key in claude.ai, with no install. That route does not cover Codex or plan limits.

You decide whether prompts are stored.

The admin chooses whether the text of prompts and replies is kept. Switched off, spend, tokens and timing, tool calls, commands and repository names are still kept.

Engineers see their own numbers, never a teammate's. Admins can read prompts while storing is on.

For engineers Docs (opens in a new tab)

Example team

OwnerMIMaya Iyer

Privacy

Store prompt and reply text

On: the text of each prompt and reply is kept with its event.

Off: new events are kept without the text. Text stored before you turn it off stays.

Prompt capture is on.

Always kept, whichever you choose

Which agent and model, and when
Tokens, cost and how long each request took
Tool calls, commands and repo names
Risk checks, such as a secret found (its kind, never the value)

Example team · November 2026

You write the code.

What you get, how to install it, exactly what is sent, and who can see it.

For engineers

You answer for the bill.

Free for one person and for small teams. Create a workspace, run one command, and the numbers start arriving.

Start free, no card

First hit	Tool	Limit	Evidence	Reset
Nov 9, 2026, 15:07	Codex	5-hour	100% recordedBlocked	16:03
Nov 5, 2026, 18:41	Codex	5-hour	100% recordedBlocked	20:12

See what used up your Claude Code or Codex limit.

Find the turns that ate your 5-hour window.

See how close you are, from every machine you use.

Then bring your team.

Who ran past their plan, and why.

From quota to pull request, feature by feature.

One command per machine.

Sign up

Run it once

Send one prompt

You decide whether prompts are stored.

You write the code.

You answer for the bill.