AI Limits FAQ

Does History update after manual refresh, automatic refresh, or both?

Both. A successful Connect, provider Refresh, header refresh button, or scheduled automatic refresh adds one normalized typed History observation containing the quota, counter or spend, and balance metrics returned by that provider. History graphs quota metrics and the balance of balance-first providers; counter or spend observations and other balances remain stored but are not graphed.

Failed, deferred, skipped, superseded, or malformed refreshes do not add a history observation. They also do not insert a zero or copy the last value forward.

Does automatic refresh require the side panel to stay open?

No. AI Limits uses a Chrome alarm and its Manifest V3 background service worker, so closing the side panel does not stop scheduled refreshes. When you reopen the panel, it reads the latest data from local extension storage.

What must be true for automatic refresh to run?

Automatic refresh must remain enabled in Settings.
At least one provider must remain connected. AI Limits removes the alarm when no providers are connected.
Chrome must still have that provider's optional permission.
For ChatGPT, Claude, Kimi, Cursor, Grok, Mistral, and Perplexity, the signed-in browser session must still be usable. API-key providers instead need a saved active key; configurable instances also need their exact instance permission.
Chrome must be running and able to run background extensions.

The alarm runs approximately every 15 minutes. Chrome scheduling, device sleep, provider backoff, and network conditions can delay a refresh, so this is not an exact sampling interval.

What happens while Chrome is closed or the computer is asleep?

AI Limits cannot refresh while Chrome is fully closed or the device is asleep. Chrome may deliver a delayed alarm after it resumes. AI Limits also checks and recreates its refresh alarm when the extension worker or Chrome starts, but it does not reconstruct observations for intervals that were missed.

If manual refresh works, should automatic refresh also work?

Usually for browser-session providers other than Kimi and Grok, as long as their permission and session remain valid, and for API-key providers while the saved key and host permission remain active. However, manual refresh bypasses scheduled backoff, while automatic refresh respects it. Kimi manual refresh may additionally perform interactive session recovery. Cursor manual refresh may additionally request Grok Bot and extra-credit JSON through an already-open Cursor page, or by briefly opening one inactive spending tab when none is open; automatic refresh never opens or injects into a page. Scheduled refresh keeps last-good page values until Grok Bot's weekly reset and explains when they could not be refreshed. Grok Connect or manual Refresh read grok.com from a page because extension-background requests are rejected; they prefer an already-open grok.com tab and may briefly open one inactive tab if none is open. Scheduled Grok refresh may inject into an already-open grok.com tab and never opens a new one.

Kimi has an additional limitation described below.

Why can Cursor Grok Bot or extra credits appear only after manual refresh?

Cursor's base monthly and on-demand usage refreshes in the background. Its Grok Bot and extra-credit dashboard endpoints reject cross-origin POSTs (Invalid origin for state-changing request), so AI Limits reads them only during Connect or an explicit manual Refresh. It prefers an already-open cursor.com tab. If none is open, it may briefly create one inactive spending tab, wait up to 10 seconds, and close only the tab it created. The extension runs a bundled exact-origin function that sends three fixed read-only dashboard requests and returns their JSON for schema validation in the extension. It does not inspect rendered content, browser storage, or cookie values directly; Chrome attaches the signed-in Cursor cookies to those fixed same-origin requests. It never activates a Cursor tab, and scheduled refresh never opens or injects into a Cursor page. Last-good Grok Bot and extra-credit values stay on the card until Grok Bot's weekly reset; the card says why they could not be refreshed.

Why does Grok need a grok.com page?

Grok's session and usage endpoints reject extension-background requests when the Origin is not https://grok.com. AI Limits therefore reads them only by running bundled exact-origin code in a grok.com page. Connect or manual Refresh prefer an already-open tab. If none is open, they may briefly create one inactive https://grok.com/ tab, wait up to 10 seconds, and close only the tab they created. Scheduled refresh may inject into an already-open grok.com tab and never opens a new one. The function does not inspect rendered content, browser storage, or cookie values directly; Chrome attaches the signed-in Grok cookies to those same-origin requests.

Why doesn't Kimi automatic refresh always work, and why can manual refresh open a background tab?

Kimi automatic refresh is best-effort and may not always work; a manual Connect or Refresh may briefly open an inactive Kimi tab in the background to recover the session.

Kimi scheduled refresh never opens a tab. An interactive Connect or Refresh may open at most one inactive temporary Kimi tab, waits up to 30 seconds for recovery, and closes only the tab it created. Recovery waits until Kimi finishes loading the page before accepting a session token, and it leaves the tab open long enough for Kimi to replace an expired token. It can use the legacy Kimi cookie or read an access_token from an already-open Kimi page. If neither is available, automatic refresh defers until a later refresh, preserves the last good data, and adds no history observation.

A manual Connect or Refresh is interactive and may use that inactive homepage tab to let Kimi refresh its own session. This difference means manual Kimi refresh can work when scheduled refresh cannot.

Why does ElevenLabs need an API key, and can I reopen the setup page after signing in?

ElevenLabs uses a user-created API key; after a successful check, AI Limits stores it locally and scheduled refresh does not open an ElevenLabs tab.

Clicking Connect ElevenLabs opens both the in-panel guide and ElevenLabs' official API-keys page. If the page asks you to sign in, finish signing in and use Open API keys page in the guide to reopen it. Create a key named AI Limits, select User → Read, and avoid generation or write permissions. ElevenLabs does not formally publish the exact mapping between that scope control and the subscription endpoint, so Validate & connect trims the candidate and checks the real read-only request; only a successful validation is saved, and AI Limits never enables broader scopes for you.

Only your explicit setup or reopen action opens the normal ElevenLabs page. After connection, manual and scheduled usage refreshes call the API in the background and do not open or inspect an ElevenLabs tab.

How is the ElevenLabs key stored, and what happens if it is rejected?

After validation, AI Limits keeps the key in a separate local credential record whose Chrome storage access is restricted to trusted extension contexts. It is not copied into usage state or History and is sent only to the ElevenLabs API for the read-only subscription request. This is extension-origin isolation, not OS-keychain encryption: someone with access to your unlocked Chrome profile, extension DevTools, or profile files may be able to inspect it.

If ElevenLabs later rejects the saved key, AI Limits stops scheduled ElevenLabs requests and shows Replace key while preserving the last good normalized usage and History. A failed replacement leaves the prior key unchanged. Disconnecting ElevenLabs, choosing Delete all local data, uninstalling the extension, or clearing extension storage deletes the saved key. External host-permission removal instead marks ElevenLabs as requiring permission and retains its local configuration, saved key, normalized usage, refresh status, and History until reconnect or explicit disconnect/delete.

What does New API support, and what URL should I enter?

AI Limits supports multiple independent New API instances. Each connection has its own normalized base URL, optional label, relay key, current usage, refresh state, replacement/rejection status, and History. It reads only that key's /api/usage/token/ response: capped keys show granted, used, and remaining quota, while unlimited keys show absolute usage without an invented percentage. It does not read account wallet, subscriptions, or admin data.

Two New API instances on the same origin share only Chrome's browser-global origin permission; they never share a relay key, label, usage state, or History. Disconnecting one instance deletes only that instance's key, configuration, usage, refresh status, and History before permission cleanup. The shared grant remains while a sibling still owns it; disconnecting the final owner removes it best-effort and keeps durable cleanup evidence for retry. Externally removing the grant instead marks every affected instance as needing permission while retaining its nonsecret configuration, saved key, normalized usage, refresh status, and History until reconnect or explicit disconnect/delete.

You can paste the site homepage, a dashboard URL, /v1, /v1/messages, or /api/usage/token/. AI Limits removes those known suffixes, preserves any deployment subpath, and validates the result through /api/status. HTTPS is required except for localhost development.

The usage request is read-only, but a relay key is not necessarily a usage-only credential: the same key may still be able to call models. Use a dedicated key and apply quota, model, IP, or other restrictions in your New API instance where appropriate. The validated URL and key are saved locally for background refresh. See Supported providers for the full boundary.

How is OpenAI different from ChatGPT?

ChatGPT uses the signed-in browser session on chatgpt.com. OpenAI uses a user-created API key against the Platform organization and developer billing surface: organization costs and completion usage first, with a fallback to legacy credit grants when organization scope is unavailable.

Does automatic refresh open provider tabs?

No. Scheduled refresh is non-interactive and never creates provider tabs. Only an interactive Kimi Connect or Refresh may create an inactive Kimi tab for session recovery and wait up to 30 seconds for the page to finish loading. An interactive Cursor Connect or Refresh may briefly create one inactive spending tab when no cursor.com tab is already open, waits up to 10 seconds, and closes only the tab it created. ElevenLabs opens its normal API-keys page only when you explicitly start setup or choose Open API keys page. Other API-key providers, including New API, do not open provider tabs for setup or refresh.

What does History store, and for how long?

History stores successful normalized quota, counter or spend, and balance observations per instance. Currency-denominated spend counters and balances are normalized usage data, not raw payment transaction history. History graphs quota metrics and the balance of balance-first providers; counter/spend observations and other balances remain stored but are not graphed. History never stores credentials, raw provider responses, or reconstructed provider history. Failed, deferred, skipped, or malformed refreshes do not add History observations; AI Limits separately stores only the latest sanitized refresh status.

Observations are retained locally for up to 30 days, subject to a 1,024-observation safety cap for each instance. Within that cap, observations from the newest 48 hours stay at collection resolution; older retained history keeps the latest observation in each UTC hour. Disconnecting an instance or deleting all local data removes its History. External permission removal retains History but pauses refresh until permission is restored or the instance is explicitly disconnected/deleted.

Why does the graph say it needs another successful refresh?

A line needs at least two observations for the selected quota window. On upgrade, AI Limits may preserve the current valid snapshot as the first observation at its original fetch time. One more successful refresh that includes the selected quota window gives the graph a second point.

Why are there gaps or separate segments in a graph?

AI Limits does not draw a continuous line across a missing quota window, a gap longer than 90 minutes, or a changed reset boundary. This avoids presenting a provider reset or a failed refresh as consumption. A window that disappears from the provider's current response is not selectable until it returns.

A balance chart breaks its line after 3 hours without a reading. A rise in the balance draws as a step up, because a top-up lands between two readings.

Does switching between Used and Left change stored History?

No. AI Limits stores one canonical used ratio. Used displays that ratio; Left displays its complement. The preference changes presentation only, not the stored observation.

What should I do before switching provider accounts?

Disconnect that provider instance in AI Limits before changing accounts. History is scoped to the provider instance and stable metric identifier, not a persistent provider account identifier. Disconnecting clears that instance's history and prevents same-plan accounts from being combined.

For data handling details, see Privacy. To report a problem, use GitHub Issues without including cookies, credentials, private usage data, or other secrets.