Anagram solver
Rearrangements of your letters that are actually English, including the multi-word ones: dormitory has no single-word anagram, but it has dirty room.
Completely free: no model, no metering, no credits, no sign-in required.
Results
How the list is ordered, and what you can do with it
Best first is a ranking, not a shuffle: it prefers answers with fewer words, with the commonest vocabulary, and with the most balanced split between those words, because a two-word answer out of everyday words is almost always the one you wanted. The other orders sort by fewest words, by longest short word, or alphabetically. A single-word anagram is resolved by one hash lookup with no search at all, so it appears instantly however long the list behind it takes.
The minimum-length filter is the most useful control here. Without it a long phrase produces thousands of answers padded out with two-letter filler; raising the floor turns a wall of results into a list worth reading and makes the search itself faster.
Results stream in as they are found and the page stays responsive throughout, because the search is a suspendable generator driven a slice at a time and it runs entirely in this tab. You can export the result list as plain text, Markdown or CSV, and keep the answers worth keeping on your SkillSafe account, where they can be searched by meaning later. Long phrases are limited only by the word-count slider and your patience.
Try one of these
Each runs the real search against the real dictionary, for nothing.
What it is costing
What these numbers mean
There is no model here and nothing to bill. What this app spends is dictionary bytes over the wire, a slice of your frame budget while a search runs, and - only if you keep an answer - a little storage. All three are shown rather than hidden.
Sign in to see how much storage your saved answers use.
Answers you kept
Optional and free. Kept answers live on your account, not this browser, so they follow you.
How it works, in one page
Every word in the dictionary is filed under its letter signature — its letters
sorted into order. Ate, eat, eta and tea all become
aet, so they are one entry, not four. Asking "is there a word made of exactly
these letters" is then a single hash lookup rather than a walk through the whole word list. That
alone answers the one-word case instantly.
Multi-word search is the hard part, and it is where naive solvers freeze the tab. The trick is to stop looking at the dictionary. When you type a phrase, one pass over the signatures in the index keeps the few hundred or few thousand that fit inside its letters, and every step after that works on that list. The dictionary is never touched again.
Then the search subtracts. Take a candidate, remove its letters, recurse on what is left. The part that makes this finish is choosing which letter to solve for next: at every step it picks the rarest remaining letter and only considers words containing it. That forces progress — some word has to use that letter — and it makes each answer come out exactly once rather than once per ordering of its words. Two prunes do the rest: a branch dies the moment it would leave a stub shorter than your minimum word length, and dies again if the letters left over could not fit in the words still allowed.
Even so, a long phrase is genuinely a lot of work, and the honest thing is not to pretend otherwise. The search is a generator that can be suspended between any two steps, and it is driven ten milliseconds at a time, one slice per animation frame. So results stream in as they are found and the page keeps responding to you while it thinks. Measured on a 35-letter phrase at six words per answer: 2,420 ms of total work, 242 slices, worst slice 10.1 ms against the 16.7 ms a 60fps frame allows.
The dictionary is SCOWL 2020.12.07, cut into four bands and shipped as four files. Turning on proper nouns fetches the fourth file once and merges it into the same index — switching a band off afterwards costs nothing, because a band is a bitmask test at search time, not a rebuild.
Questions people ask
What is a multi-word anagram?
An anagram that uses the letters of a phrase to spell several words instead of one. "Dormitory" has no single-word anagram at all, but its nine letters rearrange into "dirty room". Most solvers only look for one-word answers, which is why they report that most words have no anagram. This one searches for sets of words whose letters use the input up exactly, up to five words per answer.
How does the search avoid hanging the browser?
Three things. Words are grouped by letter signature, so "ate", "eat", "eta" and "tea" are one entry and lookup is a hash hit rather than a scan. Candidates are collected once per query rather than once per recursion step. And at each step the search picks the rarest remaining letter and only considers words containing it, which both forces progress and stops the same answer being produced twice in a different order. The search then runs as a suspendable generator sliced across animation frames: measured on a 35-letter phrase at six words per answer, 2,420 milliseconds of work split into 242 slices with a worst slice of 10.1 milliseconds against the 16.7 a 60fps frame allows.
Which dictionary does it use?
Why can I set a minimum word length?
Because without one, a long phrase produces thousands of answers padded out with "ai", "oi" and "ut". Raising the floor to three or four letters is the single most effective way to turn a wall of results into a list worth reading, and it makes the search itself much faster: a branch that can only be finished with a two-letter filler is abandoned as soon as the remaining letters drop below the floor.
Does this cost credits or need an account?
No. There is no AI model behind this app, so there is no run to meter and nothing to charge for. It is completely free and works signed out. Signing in is optional and does one thing: it lets you keep the answers you liked on your account, so they follow you to another browser and can be searched by meaning later.
Why does it not find the funny anagram I know?
Usually one of three reasons. The answer needs more words than the word-count slider allows. It contains a word shorter than the minimum length. Or it needs a band that is switched off - proper nouns and rare vocabulary are both off by default. Raising the word count is the most expensive of the three, so try the other two first.