SuperAPI AI Directory

AI Reset Tracker

AI Reset Tracker is a real-time, source-linked tracking website that aggregates the usage limit reset situations of AI coding plans such as Codex, Claude Code, GLM Coding Plan, and Higgsfield, and provides 24-hour and 48-hour reset probability predictions as well as a free API. It polls official X accounts, status pages, release notes, documentation, and community discussions, automatically classifying new information as resets, banked resets, announced resets, boosts, limit changes, or incidents, and assigns a confidence score; high-confidence official messages can be published automatically, community messages are marked as confirmed only after verification, and unconfirmed reports do not affect predictions. The website explicitly states that predictions are based on a memoryless cadence model, that numbers are only shown after at least three completed reset intervals, otherwise it displays "collecting history," and that unconfirmed reports are flagged and never move predictions. Data is available via free JSON, Atom feeds, and an MCP server, with no key required; when displaying data, the source must be credited with a link.

Open website

About this tool

AI Reset Tracker is an independent tracking project for users of AI coding tools, mainly used to record and predict the usage limit reset situations of various AI coding plans in real time. According to its official introduction, the currently covered targets include Codex, Claude Code, GLM Coding Plan, and Higgsfield. The website displays the most recent reset time of each plan, official or community messages related to resets, and probability predictions for a reset occurring within the next 24 and 48 hours.

In terms of main capabilities, the tracker continuously polls official X accounts, status pages, release notes, documentation, and community discussion threads. It uses a classification model to judge each new piece of information, identifying whether it belongs to a reset, a banked reset, an announced reset, a boost, a limit change, or an incident, and assigns a confidence score. Official reports with high confidence can be published automatically; community reports remain "unconfirmed" until corroborated by an official source or at least three independent reports. Any uncertain content enters a review queue. All entries are linked to their sources, unconfirmed reports are flagged, and they do not affect prediction values.

Regarding the prediction mechanism, the website states that confirmed resets feed into a memoryless cadence model, which is backtested daily and publishes scores compared with a baseline. Predictions require at least three completed reset intervals before numbers are given; before that, it displays "collecting history" instead of generating values out of thin air. The website also states that because reset behavior is an irregular act of goodwill rather than a fixed schedule, the 24-hour probability may remain flat during the waiting period.

In terms of usage, users can directly visit the website to view tracking cards for each plan, recent reset records, average reset intervals, and the probability range for the next reset. For developers, the website provides free JSON, Atom feeds, and an MCP server, usable without a key; if displaying data, airesettracker.com must be credited with a link. The website also states that it does not touch user accounts, and personal clock data is stored in browser local storage.

The main applicable scenario is developers and heavy users who pay attention to the frequency of AI coding plan limit resets, used to understand the historical cadence and short-term probability of resets for each service. In terms of limitations, the website explicitly states that this is an independent community project, product names belong to their ownersall; predictions are based on historical intervals rather than official schedules, community reports do not participate in predictions before confirmation; some plans may be in a "collecting history" state due to insufficient historical data.