Everything the panel can do.

Each entry gets a derived status: fresh, fetching, error, inactive, or uninitialized, colour-coded, with the endpoint name, the serialized cache key, subscriber count, and when it last updated. There is deliberately no stale badge: RTK Query has no staleness model, it evicts via keepUnusedDataFor. An entry that is fulfilled with zero subscribers reads as inactive, which is the state that actually matters, because it's about to be collected.

The counts in the header are also the filter. Click any status to narrow the list; unselected pills dim so it's obvious a filter is on. They collapse to dots and counts when the panel is narrow.

Search is fuzzy by default, so lpf finds listPostsFlaky. Toggle .* for regular expressions when you need precision. A pattern that doesn't compile leaves the list unfiltered and marks the toggle instead of blanking the panel while you're mid-keystroke.

Order by last updated, endpoint name, or status severity, ascending or descending. The default, most recently updated first, is usually what you want when something just changed.

Select an entry for its endpoint, type (query or infinitequery), subscriber count, timings, and duration. Provided tags render as chips that jump straight to the Tags tab. Subscriber counts carry a note that they can lag ~500ms: that's RTK Query syncing subscriptions on a throttled timer, not a bug in the panel.

A cache entry only remembers its latest request. This lists every request that targeted it: each refetch, its outcome, duration, and whether it was forced. It's how you answer "has this refetched, how often, and did any of them fail?"

Refetch forces a fresh request. Invalidate invalidates everything the entry provides. Remove drops it. Reset API clears the whole cache, behind a confirmation. Refetch needs the real api object, so it's disabled with an explanation if you didn't pass apis.

The tree never serializes a whole value up front, which is the usual way a devtool freezes on a large cache. Every node has a copy button that serializes on demand. Large collections split into chunks of 100 rather than rendering fifty thousand rows.

Map, Set, BigInt, Date, functions, symbols and circular references all render sensibly. Error instances show name, message and stack. Without that they'd appear as an empty {}, since those fields aren't enumerable.

Each detail pane ends with the complete derived entry, so nothing is hidden behind the curated rows above it, including the true cache key, raw status, and request id.

RTK Query doesn't keep mutation arguments in state, so the panel recovers them by correlating against the timeline. You get the arguments, the result or error, timings, duration, and a Remove action. Same search and sort controls as Queries.

Tags group by type, then by id, and expand to the exact cache entries each one touches. Click a cache key to jump to it in Queries. Invalidate a single id or the whole type from here. Tags provided without an id show as none rather than RTK's internal sentinel.

Each entry shows the endpoint, whether it was a query, mutation or infinite query, its outcome and how long it took. Deduped requests are marked skipped rather than counted as failures. Pause capture when you want a stable view, or clear it. The buffer holds the last 500 events by default, configurable via maxTimelineEntries.

Fastest, median, average and slowest per endpoint, ordered slowest first, with error counts. Cache state can't tell you this, because it only holds the latest request's timestamps, so "which endpoint is slow, and how often does it fail?" is only answerable from the log. Skipped requests are excluded so they can't drag the numbers toward zero.

Toggle offline or blur the tab without touching your network or your code. These dispatch RTK Query's own global actions, so refetchOnReconnect and refetchOnFocus fire for real when you switch back. They aren't cosmetic flags.

keepUnusedDataFor, invalidationBehavior, and the three refetchOn* flags, plus a count of cached queries, mutations and subscribers. These explain cache behaviour that otherwise looks arbitrary. Collapsed by default so it costs no space.

If RTK Query reports middlewareRegistered: "conflict", meaning the same api's middleware registered twice, usually a duplicated module or bad HMR, the panel shows a persistent warning. Caching and invalidation misbehave silently in that state, and no other tool reports it.

Every RTK Query api in your store is found by inspecting state, with no configuration. An api selector appears in each tab once there's more than one.

Queries, mutations and the timeline are all virtualized, so hundreds of cache entries stay responsive.

Active tab, filters, search terms, sort order and expanded sections persist across reloads, namespaced so they can't collide with your app's own storage.

Both the middleware and the plugin become no-ops when process.env.NODE_ENV === "production", so bundlers drop the panel entirely. Leaving them wired into your store and app is safe.

The panel follows the devtools shell's theme. Tabs are wired with proper roles and labelling, toggles expose their pressed state, and the data explorer is operable from the keyboard.

Supports @reduxjs/toolkit from 2.0.0 up, including the internal change in 2.6.2 that reshaped the tag index. It's handled transparently, so the panel doesn't care which side of that line you're on.