One cat. Three kinds of space.
- 1 Katze pro Farbe
- 1 Katze pro Reihe & Spalte
- Katzen nie benachbart
One cat per color
Every colored region holds exactly one cat, whatever its shape.
One cat per row & column
A placed cat claims its whole row and its whole column.
Cats never touch
Not even at the corners — every neighbour of a cat is off limits.
Paws first, cats last
Tap once to paw-mark a cell that cannot hold a cat; place a cat only when it is forced. Stuck? A hint names the rule that unlocks the next move.
Every valid board follows four promises.
Exactly one per row
Each horizontal line contains one cat — no more and no fewer. A row holding two cats is invalid even when every other constraint is satisfied. A row holding none means the board is unfinished, not wrong.
Exactly one per column
Each vertical line contains one cat. Together with the row rule this fixes the population of the board: an N×N grid holds exactly N cats, one per row and one per column, with no line ever shared.
Internalize this early, because it turns counting into a tool. If you have placed six cats on an 8×8 board, exactly two remain, and they must live in the two rows and two columns still standing empty. Late-game positions are often finished by counting rather than scanning.
Exactly one per colored region
The board contains exactly N connected regions, and each holds one cat regardless of its shape or area. A region of two cells and a region of eleven cells carry the same single cat, which is why small regions are worth far more attention than large ones.
Connected means orthogonally connected: every cell in a region can be reached from every other by moving up, down, left, or right without leaving the color. Two same-colored blocks that meet only at a corner are two regions, not one — a distinction worth tracing carefully on a crowded board, and the single most common thing to misread in a screenshot.
Cats never touch
Two cats cannot share an edge or a corner. Every placed cat forbids up to eight surrounding cells, and this is the only rule that constrains cells outside the cat’s own row, column, and region.
That makes it the most productive rule on the board. The other three describe lines and colors you were already reading. The no-touch rule reaches diagonally into regions you were not.
Every valid puzzle has exactly one solution
A board admitting two different arrangements is not a harder puzzle — it is a broken one. Every puzzle generated here is verified to have a unique solution before it is served, and that solution is always reachable through certain moves alone. You should never need to guess and check.
Board sizes
The solver accepts boards from 4×4 through 12×12. The generator makes boards from 5×5 through 12×12, because a 4×4 grid rarely leaves room for an interesting region layout. Difficulty is not a function of size alone: a well-shaped 7×7 can be harder than a loose 10×10.
Starter cats
Some boards arrive with cats already placed. They obey the same rules and are simply deductions the puzzle grants you for free. If starter cats conflict with one another — sharing a row, a column, a region, or a corner — the board is invalid, and nothing after that point is worth solving.
Imported boards
A screenshot or hand-drawn board must have N connected regions and non-conflicting starter cats. If it is invalid, the solver reports no solution or multiple solutions instead of inventing an answer.
Both answers are useful diagnostics. No solution usually means a region boundary was misread and one region got split in two. Multiple solutions usually means a boundary was missed and two regions were merged into one. In either case, correcting the region outline in the editor is faster than re-taking the screenshot.
Make the impossible cells do the work.
Finish every consequence
After placing a cat, mark its row, its column, its region, and every adjacent cell. One missing mark can hide the next certain move, and a board that looks impossible is usually a board with an unfinished sweep somewhere behind you.
When you are stuck, do not look for something new first. Go back to the last two or three cats you placed and re-trace their consequences. The missing deduction is there more often than not.
Lock regions onto lines
If every remaining cell in a region lies on one row, that region’s cat must occupy that row — even though you cannot yet say which cell. Every other cell in that row therefore becomes impossible, including cells in completely different regions.
This is the highest-value pattern in the game, and it costs nothing to check. After each placement, glance at the small regions and ask whether their survivors have collapsed onto a single line.
And lock lines onto regions
The same argument runs backwards. If every remaining candidate in a row sits inside one region, then that region’s cat is the row’s cat. Every other cell of that region — anywhere else on the board — can be eliminated. Players who only read the pattern in one direction leave half its value on the table.
Compare candidate sets
If two rows are limited to the same two columns, those columns are reserved for that pair. This often unlocks a third row without ever deciding which of the two cats goes where.
The pattern scales. Three rows confined to the same three columns reserve all three, and the same reasoning applies to columns confined to rows. You are not identifying cats — you are identifying territory that is already spoken for.
Count what is left
Late in a board, stop scanning and start counting. Cats remaining, rows remaining, columns remaining, and regions remaining are all the same number, and matching them against each other collapses positions that look wide open. If three cats remain but only three regions are unresolved, each of those regions takes exactly one, and any cell of theirs that cannot serve is finished.
Mind the geometry of edges
The no-touch rule is weaker at the border. A cat in the middle of the board forbids eight neighbors; a cat in a corner forbids three. Regions that hug an edge are therefore looser than their size suggests, and central regions are tighter. When two candidate placements look equally plausible, the one nearer the middle usually rules out more and is worth testing first.
Use contradiction sparingly
On expert boards, assume one candidate holds a cat and follow only certain consequences. If the assumption produces a conflict, you can safely eliminate that candidate. The full replay labels this technique honestly as bounded contradiction.
Keep the chain short and keep it certain. A contradiction you followed for three forced moves is a proof. A contradiction you followed for fifteen speculative moves is a guess wearing a proof’s clothing, and it is how solvable boards get abandoned.
Mistakes worth knowing
Three errors account for most stuck boards: forgetting the diagonal neighbors when sweeping a placed cat, treating two corner-touching blocks of the same color as one region, and marking a cell impossible for a reason you never actually verified. The first two are mechanical. The third is the dangerous one, because a single unjustified paw makes the rest of the board lie to you.
Good things to know before your first cat.
Does my screenshot leave the browser?
No. Recognition runs locally in browser memory. There is no screenshot upload endpoint, and the original image is discarded when you close the page or choose another image.
What does it cost?
Nothing. Screenshot recognition is limited to five boards per UTC day because it is the only part that is expensive to run; watching one optional ad adds three more. Manual boards, hints, complete answers, the step-by-step walkthrough, and generated puzzles are unlimited.
Which sizes work?
The solver accepts boards from 4×4 through 12×12. The generator creates boards from 5×5 through 12×12.
Will every imported board have a full explanation?
No. The answer finder can solve boards that require techniques outside the explanation engine. When that happens, it returns the answer and clearly marks the replay as incomplete.
Is this official?
No. This is an independent, unofficial helper built with original code, artwork, and generated puzzles. It is not affiliated with or endorsed by Oakever Games.