CODEX RESCUE / THE INDEPENDENT HELP DESK

Codex errors?
Let’s troubleshoot.

From a broken stream to a missing tool.
Find the next step for your Codex error.

What went wrong?

ERROR LOOKUP

Your error stays in this browser.

Try an example:
Built for CLI, desktop & IDE usersExplore the troubleshooting library

START WITH THE QUESTION YOU HAVE

Choose the right Codex help

A service incident, a personal usage limit and a local error can look similar while asking for different next steps.

01 / TROUBLESHOOT

Codex errors

Start here when you have a literal error message or a repeatable symptom in a particular client. The guide finder above uses the words you enter, plus the client and system you select, to suggest a relevant page. Each guide separates documented behavior from a public report and names the first check before any deeper change.

If a command is missing, test the shell and install path. If a session stalls, preserve the task and identify the stage that stopped. A single “Codex is broken” diagnosis is less useful than one exact line from the failing step.

Browse error guides →
02 / CHECK YOUR ACCOUNT

Limits & resets

A usage notice is about an allowance shown for your account or workspace, not proof that the Codex service is down. Identify which allowance was exhausted and use the reset time actually displayed to you. Our usage guide points to the account view and CLI check documented by OpenAI.

For planning, keep the short window, weekly window, available credits and any banked reset distinct. A public announcement can be worth reading, but it does not disclose your personal balance or a guaranteed reset time. Save a countdown only from a time you can see in your own account.

Understand your limit →
03 / VERIFY THE SERVICE

Codex status

Before reinstalling or changing credentials, check the official incident feed and the specific Codex component relevant to your client. A reported outage can explain widespread failures, while a healthy component report does not rule out a local network, account or configuration problem. We show unknown when the source is stale or unavailable.

See incidents and component details →

SYMPTOM → NEXT PAGE

Common Codex problems at a glance

Use the exact words on your screen where possible. The rows below are navigation clues, not claims that every appearance of the phrase has one cause. Select the page that matches the failing operation; if the error is different, use the lookup at the top. A status check is still useful before a local repair, especially when failures begin suddenly across several clients.

Error or symptomWhy this pageGuide
stream disconnected before completionA response stopped mid-turn; save the unfinished task and inspect the connection path.Stream disconnected →
Codex sign-in or callback failsSeparate the browser step, local callback and account exchange before changing settings.Sign-in stages →
codex: command not foundThe shell cannot resolve a CLI executable; identify the install environment and PATH.Command not found →
Codex update failed or version stayed oldCheck the installation method and executable path before installing a second copy.Update or stale runtime →
Windows setup didn’t finish • helper_failedRecord the UAC outcome and first helper error rather than repeating the setup loop.Windows sandbox setup →
config.toml cannot load or app freezes after editUse the parser path and line to distinguish a duplicate table from a bad escape.Configuration syntax →
MCP startup failedFind the failing server and transport before assuming the entire Codex client is down.MCP startup →
MCP tools are not showingThe server may be configured yet its tools are missing in the active client session.Missing MCP tools →
Codex session is stuckPreserve the current task and distinguish a stalled turn from a service incident.Stuck session →
Error creating task: TimeoutIdentify the task-creation stage; it is different from a message-submit timeout.Task creation timeout →

The first row is about a response stream; the task-creation row is about a task that never starts. Likewise, an MCP server failing to start is different from a server that starts but exposes no tools. These distinctions let the guide give you a relevant first action rather than a general reinstall suggestion.

WHEN THE MESSAGE SAYS LIMIT

How Codex usage and resets fit together

OpenAI's Codex plan guidance says usage limits vary by plan and directs you to Settings or the usage dashboard to inspect the exhausted allowance, credit balance and any reset time shown. An active Codex CLI session can also show usage through /status. That is the right starting point: this website has no access to your subscription, workspace or remaining quota.

If you see a usage limit reached message, first identify whether it is a short window, a weekly allowance, credits or another workspace control. Do not try to derive a universal reset time from the clock or from someone else's screenshot. The 5-hour versus weekly guide explains why two displayed windows can have different remaining amounts and different dates. The values and reset times shown in your account remain the source of truth for your next action.

Some eligible accounts may see a banked reset, credit purchase, an upgrade or a wait-until-reset option. OpenAI says to check the options available in the limit notice; it does not promise that every account has every choice. A banked reset and an automatic reset are not interchangeable actions. Verify the scope and the account view after applying any option. Our personal timer can remind you of a time you entered, but it cannot restore access or confirm that an account-side reset happened.

CHECK THE SERVICE BEFORE A LOCAL FIX

Is Codex working right now?

The live Codex status page reads official component and incident data and keeps the last known event visible when a refresh fails. It distinguishes the CLI, web, API, VS Code extension and desktop component. Look at the component that matches the client you are using: a desktop incident may not describe a CLI problem, and an API incident is not automatically a ChatGPT-plan limit.

Check the event's status, affected components and report time before drawing a conclusion. An ongoing official incident can justify waiting rather than changing a working installation. A resolved incident can still be relevant to the time your error began. If the data is old or unavailable, the page says so rather than pretending that there is no outage. You can open the linked official incident for its own updates.

If no relevant public incident appears, continue with the exact local message: test the failing client, account and network path. The outage-versus-local-error checklist walks through that distinction without treating one status indicator as a diagnosis. Preserve your task and change one thing at a time so you can tell which observation actually helped.

How to use Codex Rescue without losing the original clue

Paste the shortest complete error line into the lookup above and choose the client and operating system where it happened. A Windows desktop setup failure, a CLI command-path error and an IDE extension warning can share words but require different checks. The search result is a starting point, not a diagnosis made from private account data; we do not read your local logs or configuration through this page.

On the suggested guide, compare the verbatim error block and the “When it happens” scope with your own case. Follow the numbered checks in order, keeping any branch conditions. Record what changed after each step. If a guide cites an issue report, read it as evidence that someone observed that behavior, not as proof of a universal repair or a fixed release. If the guide cites an official document, use its current version before changing an installer, sandbox or account setting.

When the message does not match, use the full error library or the five category lists below. For an exhausted allowance, move to limits and resets rather than treating a quota message as a software crash. For a possible incident, check service status first. Keep session history and configuration backups before any destructive troubleshooting, and share only redacted error details if you need help.

Common Codex questions

Why does Codex say I've hit my usage limit?

The allowance available to your account or workspace may be exhausted. Check the usage notice and your account view to identify which allowance is affected before treating it as a local error. Identify the usage limit you reached →

How do I check my remaining Codex usage?

Open Settings or your Codex usage dashboard. In an active Codex CLI session, /status is a documented check. Plan allowances, workspace credits and API-key billing can refer to different balances. Find your Codex usage dashboard →

Is Codex down right now?

Check the latest official Codex component reports and incident records. An operational report does not rule out an account, network or local setup problem; a stale or failed check is shown as unknown. View live Codex service status →

What's the difference between a banked reset and an automatic reset?

A banked reset is available in your account until you apply it. An automatic reset is applied to the limits covered by its announcement. Public announcements do not establish that your own account is eligible. Compare banked and automatic resets →

Does Codex usage reset every 5 hours or every week?

These are separate usage windows, rather than two names for the same allowance. Identify the exhausted window and read the reset time your account shows. A full banked reset can refresh both and change the weekly reset date. Understand 5-hour and weekly usage windows →

THE TROUBLESHOOTING LIBRARY

Browse Codex errors by category

Browse all guides

A PRACTICAL NEXT STEP

Recognize the problem?

Sources and scope in every guide
Last updated: