
Browse public reset records and their sources, distinguish quota resets from credit grants, and opt into email alerts.
Used this tool? Rate it
Used this tool? Rate it
From a prompt to a deck, site or doc
DeepSeek AI models for chat and API
One Gateway, Every Model, Compensated Failures
One workspace for every AI model
One workspace where your team and its AI agents think together
One gateway for text and media models
AI assistant for structured work
Plan, build, and refine work with an AI agent
Codex Reset is a browser-based tracker for public OpenAI Codex reset events, with email alerts you opt into. Its public event history can put a report about extra usage next to the source that prompted it. A public signal does not confirm your personal quota.
The service identifies itself as independent of OpenAI, with no OpenAI affiliation, agency or endorsement. The operator uses Codex as the name of the monitored subject and says rights remain with their owners.
The site distinguishes when a source post appeared, when it was first collected and when the current record version was published. A change to the publication time can therefore reflect a revision to the tracker entry rather than another grant or reset. On a record that shows a third-party observation time and a later first-observed time, the second clock describes collection. Neither time alone establishes when an allowance changed.
A date-only source remains a date, an unspecified time remains unknown and a stated interval stays a range. The tracker does not convert a report that specifies only a day into an event exactly at midnight.
The alert guide describes rule-based 24- and 48-hour estimates that give no current rating when evidence is missing or outdated. A high rating in that guide cannot create a completed reset record. Neither stands in for your individual schedule.
The Activity page currently marks third-party forecasts as view-only and says estimate-change emails are paused. The help guide describes likelihood changes as a separate consent choice. That description of consent does not mean delivery is active.
An account usage limit, a workspace restriction, a network problem and a service incident can interrupt work for different reasons. The public Codex reset tracker cannot diagnose any of them for a specific account. Someone who has one interrupted session and continuing access in another workspace may need the sign-in and limit context more than the public history. A teammate seeing the same announcement may also have a different account state.
The signed-in OpenAI usage dashboard is an account view for checking current limits. For redeeming an available reset, OpenAI documents /usage in Codex CLI and gives the desktop app or web as alternatives if that CLI version does not show the redemption option. A public record does not read these views. A mismatch between a report and the allowance that appears after sign-in can reflect the offer's scope, the selected workspace or timing of the update. Only the account can show which limit applies to the session that was interrupted.
OpenAI describes a banked reset as a one-time opportunity saved for later redemption, so receiving it does not restore current usage. For example, a banked-reset announcement tells someone to distinguish the availability of a credit from the need to apply it. These are two separate steps in the issuer's product.
OpenAI says an automatic reset applies to the accounts and allowances selected for the offer. The word global on an announcement therefore does not settle whether one account is included or whether all of its windows refresh. A purchased reset, where OpenAI offers it to a selected Plus or Pro personal account, is paid and used immediately. A subscription at one of those tiers is not itself evidence of a paid offer.
OpenAI distinguishes these one-time resets from scheduled usage refreshes: they do not permanently raise plan limits or add usage credits. Comparing the age of a news report with the length of a scheduled window would put two different clocks on the same countdown.
One April 2026 entry attributes a reset for paid plans to the GPT-5.5 announcement and explicitly labels it as history rather than a current reset notice. A separate September announcement excludes Free and has no linked completion evidence in its current version. The September record says it was published by an operator. These records show why the eligible plans, publication method and completion status are distinct fields.
The browser interface offers English and Chinese. The public activity and history surfaces contain records to read online. Email is a separate delivery channel for opt-in notifications. Reading a public event does not require using the signup form.
The site's help uses labelled UTC timestamps for public records and applies the selected IANA time zone, including daylight saving, to email times and quiet hours. A public time label and a time printed in an email can identify the same instant. In a team spread across multiple zones, the displayed dates may differ even when everyone is discussing the same announcement.
The privacy notice lists Cloudflare for hosting and database services and Resend for sign-in and notification mail. A subscriber can still read a source record while waiting to learn whether a notification was delivered.
| Access path | What it covers | Condition to distinguish |
|---|---|---|
| Public event pages | Read the published records and follow the source trail. | Public records remain available after unsubscribing. |
| Email alert setup | Verify the destination and manage which public events are sent. | The form uses email verification without a password; choices are made after verification. |
| Trial or paid access | Receive access on the terms disclosed for the service. | Any trial or paid access follows dates shown in the account and terms given before purchase. |
No public price amount or tier table appeared on the reviewed account entry, help and terms pages. That is a boundary on what can be stated about a current offer, rather than a claim that an offer costs nothing. The terms describe any service fee as payment for the information and alert service, separate from OpenAI quota and any ChatGPT plan. The terms say there is no automatic renewal in phase one. That statement describes the named phase, so it does not settle the renewal conditions of every future offering. For a manually processed payment, the terms distinguish amount, scope, access dates and refund terms as details to confirm.
Codex Resets offers a separate Codex-focused announcement history, which can be a fit when the aim is to browse the original public posts or reuse public records. Its operator says the archive watches public Codex reset announcements from @thsottiaux. It offers public data access through an API and an MCP server. It also points to an unofficial community-built iOS app. The archive is also an upstream source for Codex Reset. Seeing its data on both surfaces is therefore not two independent reports of the same event.
AI Limit Reset Tracker, also called LimitReset, is a fit for a cross-product workflow that needs both public reports and personal time entries. LimitReset offers a personal reset timeline across tools in which the user enters times and the browser stores them. The user-entered schedule is separate from the events the public board tracks. Its vendor lists browser push without an account. Its email option is described as double opt-in, with one email per reset and a one-click unsubscribe mechanism.
Codex Reset follows selected public sources without covering every OpenAI announcement or inspecting personal accounts. The source page now lists the Codex Resets API as supplying its archive, schedules and community observations, with direct X API collection disabled. An original link can appear in a record supplied by that API. Its presence does not show that this tracker independently observed the original post. The upstream archive does not contain every post and can change through omissions or source edits.
The collection interval describes how the source is configured to be checked; it does not guarantee a new event or email within that interval. A partial or failed scan does not advance the last complete-scan time; source access, rate limits and budget pauses can affect coverage. Even if the timestamp is current, a completed scan can simply mean the currently available data yielded no eligible new entry.
The terms do not guarantee full coverage, delivery at a fixed time or the occurrence of a reset. A public source, mail sender or receiving server can also introduce delays. The sequence matters when reading an old message while the revision history has moved on. The operator says a message already in transit cannot be recalled. An email can therefore arrive after its underlying record was revised.
The privacy notice states that the tracker does not connect to your OpenAI account or obtain its password, API key, conversations or personal quota. It uses the supplied email address for sign-in codes and requested messages. The notice describes verification, session and preference records, and says sign-in security may log IP and browser information. Stored preferences include the event types, selected language, time zone, plan filters, quiet hours and consent choices. These are account records for the alert service, not measurements of the subscriber's work.
The notice says public-page analytics and replay, including the listed Microsoft Clarity replay and heatmaps, start only after consent and that declining leaves sign-in and alerts available. The operator says it does not sell subscription email lists.
Turning alerts off stops future notifications without automatically deleting the account, delivery records or audit history. The operator handles access, correction and deletion requests; the notice says there is no self-service deletion tool yet. The privacy notice says unmatched delivery receipts are removed after 30 days. The stated period concerns unmatched receipts. The operator describes other records as being retained as needed for security, service, corrections or disputes, rather than attaching the same period to every record.
The public entry records what a source announced or reported. Eligibility can differ for each offer. A banked grant can appear as an available action without changing current usage, whereas a scheduled refresh follows the account's own limits. The signed-in account identifies which allowance and recovery time apply.
Test messages are verification messages that bypass quiet hours. They check delivery to the verified inbox. A successful test does not indicate that a live event met the delivery rules or that any reset occurred.
A card may represent an earlier version. The current event detail gives the latest status and retains the older versions. Comparing the revision and source helps distinguish a correction from a new public event.
The privacy notice treats unsubscribing, delivery history and deletion separately. Turning off alerts affects future eligible messages and does not automatically delete the other records. A data request goes to the operator.