No more decision fatigue. Point it at your folder, ten options or ten thousand, and Juxtapose works out which two to ask you about, side by side.
Desktop app · runs on your own machine · Buy once, no subscription
Payment handled by the App Store or Microsoft Store
You have many plausible options and information about each one scattered across folders, tabs, PDFs and messages. Every tool you reach for was built to hold something else.
Keep your thinking in one long stream. Useful for remembering, awkward for choosing between alternatives.
Put options into rows and attributes into columns. Useful for filtering, awkward when the evidence is visual, audible or long-form.
Keep every source file intact, but not the decision. You still open candidates one by one and compare from memory.
Images, audio, video, PDFs, documents, subfolders. A folder per option, or a file per option, or one brief that describes them all.
Each option arrives as a candidate with its media attached and its numbers pulled out, every one traceable back to the document it came from.
Two options, properly presented, one question. Juxtapose works out what to ask next so you answer as few as possible.
A ranking derived from your own answers, with the contradictions surfaced rather than smoothed over, and an export when you are done.
Two hundred applicants, each with a CV, a portfolio, a take-home and interview notes. Shortlist without every panel member reading all two hundred.
Nine vendor folders of quotes, specifications, contracts and correspondence. Compare on what is actually in the documents, with every figure traceable.
Forty pieces from a dozen suppliers, with prices, dimensions and lead times spread across PDFs. Present a short list with the reasoning intact.
Three hundred creative variants. Rank on hook and clarity with names hidden, so the client's favourite does not win because it is the client's.
Takes, mixes, masters, grades. Level-matched and blind, because a take that is a decibel louder wins for reasons that have nothing to do with the take.
Thirty listings with photographs, floorplans, surveys and your own notes. Decide once, properly, and keep the reasoning for the argument later.
It turns up just as often in tender and RFP evaluation, agency and supplier selection, architecture schemes, product and UX concepts, brand identities, campaign creative, roadmap prioritisation, investment screening, literature reviews, grant and conference refereeing, model and prompt evaluation, dataset curation, casting, photo selects, game art, event venues, universities, job offers, cars, holidays, and anywhere else the shortlist is too long.
Runs locally. Your files stay on your machine. You can answer yourself, split the pool across a team, or connect a local MCP model as another judge. A desktop app, bought once.
"How good is this, one to ten?" is a question almost nobody answers consistently, and the number means something different by the time you reach option twelve. "Which of these two?" is a question people answer well.
So Juxtapose asks just that, and works out everything your answers already settled. Prefer A to B and B to C, and it will not ask you about A and C.
Every comparison is chosen for maximum information. Juxtapose asks the pairs that narrow the field whichever way you answer, rather than repeatedly comparing everything against the current frontrunner.
Prefer A to B, B to C, then C to A? Juxtapose keeps each answer, shows the smallest knot, and leaves a genuine contradiction visible instead of quietly rewriting it.
Both options get the same frame width, without a crop or display-size advantage. Audio and video can be level-matched on request, and blind mode hides names so the take is what you judge.
Numbers, dates and fields stay linked to the passage or exact region of the scanned page they came from. If the source cannot ground a value, it stays blank until you correct it.
Reads images, audio, video, PDF, DOCX, RTF, HTML, CSV and JSON. Related files stay attached to their candidate.
One licence per seat, bought once, and the trial is the whole thing rather than a crippled version of it.
7-day free trial
7-day free trial
Payment handled by the App Store or Microsoft Store
The engine never knows what a candidate is. It only ever asks which of two won, and works out the rest. That means whoever answers is a slot, and the slot takes more than one kind of occupant.
You answer the questions. Juxtapose asks as few as it can and tells you when your answers disagree with each other.
Split the pool. Everyone receives only the comparisons they are best placed to judge. Their attributed answers contribute to the same decision. Nobody has to see everything.
State the criterion and let an AI model answer the comparisons, at a scale no person would attempt. The maths underneath does not change, which is why this is a slot and not a feature.
One Juxtapose instance coordinates the decision. The others evaluate only the comparisons they are given, with evidence supplied for the pair in front of them.
Hosting is the coordination role for one decision. Any Juxtapose instance can take it, and the same app can host one decision while judging in another.
Judges receive comparison jobs, not the whole decision.
Juxtapose can connect a local MCP client as another judge. It sees only the comparisons and evidence it is given, and its answers stay attributed to that model.
A designer judges images, a copywriter judges text, an AI judges another slice, all contributing to one decision graph without centralising the files.
No. There is no cloud your files are uploaded to for processing, and no account to create. Everything works locally. If you explicitly connect another Juxtapose machine as an evaluator, evidence can be sent directly to that machine when it needs a file it does not already have.
Time, not capability. Seven days of the whole application, so a real decision fits inside the trial.
It never invents a preference. Every ranking is derived from answers you actually gave, and it can show you which ones produced any given result.
Answer again. The later answer wins, the earlier one stays in the history, and the ranking is recomputed. Nothing is lost and nothing is quietly overwritten.
No. Point it at your files and correct what it got wrong, which is a much smaller job than describing your decision to a database before you start.
Yes. When a document describes several options, you can provide a list of their names. Juxtapose uses it to match headings and filenames, while the source text stays attached as evidence.
I'm Rohith. Over a decade building production systems: four years at Meta on Workplace's billing and monetization, a senior role at HubSpot, backend services at Amazon. I have also written two books about how computers actually work.
Juxtapose exists because every tool I reached for modelled the wrong thing. A note is a document. A sheet is a grid. Neither of them is a set of alternatives you have to choose between, and I kept losing good options in the gap.
It is built the way I would want somebody to build software I was going to trust with a decision. The ranking engine is deterministic and tested against its own theoretical bound. Where the design was wrong, it was measured and changed rather than defended.
Point it at your folder and answer a few questions. An order comes out the other end.
