Straight answers

One tap saves it as a working formula, ready to weigh as a batch, and the dose is corrected for the strength of the bottle you actually own: if your only linalool is a 10% dilution, the weighed mass is multiplied up so the formula is true to what is in the bottle.

It refines. In AI Studio the conversation is kept, so "warmer, drop the aldehydes" re-runs against the previous answer, and the model's own open questions come back as one-tap prompts. From a formula you already have, it reads that formula, gives advice, and a second tap turns the advice into a complete revised version with every unchanged line kept at its share. If a returned formula fails the server's checks it is re-asked once with the exact faults listed, and it is never shipped unverified.

It rebuilds. On any community post, "Rebuild from my shelf with AI" hands that formula to the studio with the shelf-only constraint on, and what comes back is the same idea recomposed from bottles you actually own. It is deliberately not a clone, because a post names materials you may not have. It works the same way from a PDF or a pasted formula. Taken together this is probably the most advanced AI in any formulation tool today. We say probably because nobody has measured every tool, and this page does not claim what it cannot check.

What it does not do: it never edits a formula in place, so a revision is always a new copy you look at before saving. It does not browse the internet. It cannot smell anything, has no feedback from your own trials, and does not learn from your results. And it is not free: a new formula costs 10 credits. Questions use no credits on an AI plan, and direct edits use none at all.

Two tools are explicitly shelf-bound. Substitutes searches your own shelf with role as a hard gate; where no single bottle fits it rebuilds the target from a pair or a trio with a fitted ratio, and it returns an honest no-match when nothing is close. And AI Studio is grounded on your real shelf: with shelf-only on it cannot name a material you do not own, and a returned formula that fails that check on the server is corrected or refused rather than shipped.

Your shelf comes in the same way formulas do: import from PDF, CSV, TSV, JSON or pasted text, matched against 5,888 supplier and trade synonyms so the name your supplier printed finds the material we already describe. Stock, prices and suppliers ride along, and your shelf exports back out as one CSV at every tier, free included.

The arithmetic honors what is really in each bottle: every dose is corrected for the strength it is weighed at, so a line weighed from a 10% dilution counts for a tenth of the same gram taken neat, in the formula, the scent profile and the compliance checking alike. On any community post, one button rebuilds that formula from your own shelf with AI: the same idea recomposed from bottles you actually own, deliberately not a clone, because a post names materials you may not have. The formula count is nearly the only limit here: the free tier holds an unlimited materials shelf and every tool but four, and the one paid version that lifts the formula ceiling can be rented by the month or bought outright. Paid features include color PDF themes, formula version history, bottle labels and the sale-ready pack.

The catalog helps you start: hundreds of materials carry a typical working dilution, so a bottle arrives with a sensible strength instead of a blank field. And the correction runs on what you own, not on an assumption: if your only linalool is a 10% dilution, the weighed mass is multiplied up so the formula stays true to what is in the bottle.

Concentration is a derived, adjustable quantity rather than a hand-recalculation: change the batch size or the target strength and the alcohol moves with it. Making a dilution is a first-class event in stock too: "dilution created" is one of the typed movements in the append-only ledger, so the new bottle's balance is tracked the same way a purchase is.

This matters because juggling dilutions at 1%, 10% and 50% is the most common way a spreadsheet goes quietly wrong: solvent math is easy to get almost right. Here it is one calculation, and it is free at every tier; the paid version adds unlimited formulas. What that costs.

It has been checked against the official book, line by line. On 2026-08-16 the compiled IFRA Standards up to and including the 51st Amendment were downloaded from IFRA's own CDN (263 standards over 726 pages), and every entry in the app was diffed against it by CAS number. 252 of 252 comparable entries matched the published Category 4 figure exactly, with 0 disagreements. Of the rest, 8 are specification standards that publish no concentration limit at all and are correctly blank here, and 31 are class standards and constituent-driven naturals that IFRA does not key to a single CAS. The script that does it ships in the repository, so the check can be run again against the next amendment rather than taken on trust.

Three things are worth saying plainly. It is Category 4 only: this is an app for fine fragrance, and a candle or a body lotion is a different category with different numbers. It moves when a new version ships, not on its own, so it names the amendment rather than implying it tracks the current one. And it is indicative: a working figure to compose against, not a certificate. Verify against the official IFRA documents before you sell. The app stamps every compliance sheet exactly that way.

The coverage is honest too. 314 materials in the catalog are flagged restricted and 267 of them reach a standard. For the rest the app says "restricted, check the current amendment" and gives no number, because a number invented from memory is worse than a blank. The limits, material by material, are at IFRA limits by material.

The composition, since a count on its own says little: 501 naturals and 1,710 aroma chemicals, isolates and compounded bases. 1,642 carry a CAS number; the rest are mostly commercial bases, which have no single CAS to give. 739 carry an estimated constituent table. All 2,211 carry a written odor description and an impact figure, and 2,172 carry an olfactive family with percentage dominance; the other 39 are solvents and diluents, which have no smell to file. Every material resolves to a substantivity figure, 898 of them from a sourced value and the rest estimated from family and band.

You do not have to take any of this on trust: the whole catalog is free to browse without an account, one page per material. Open a few and count.

We say that plainly because the alternative is worse. There is no separate switch: an account and synced data are the same thing. If you would rather nothing left the device, do not sign in. The community and AI Studio are what you give up.

Publishing is always a deliberate act and never a default, and your stock levels, costs, suppliers and private notes are never part of anything you publish. AI Studio sends your brief and the relevant part of your shelf to our server and on to our AI provider to compose the answer, and that content is used only to answer your request. We do not sell data or share your formulas with advertising services. See our privacy policy for analytics details.

You can take everything with you at any time: a full backup exports as a ZIP holding backup.json, your library as CSV, every formula as both CSV and JSON, and your batches. You can delete your account and every piece of cloud data from inside the app.

What goes in. Strip life is how long a material lasts on a blotter, and it is the largest input: the drydown is exponential decay against it, and longevity is where the summed curve falls under the perception floor. Impact is evaporation flux times potency plus olfactory bloom. Every dose is corrected for the strength of the bottle it is actually weighed from, so a 10% dilution contributes a tenth.

Where the confidence comes from. A material's strip life is either a measured pin, which somebody timed on paper, or an estimate from its family and its tier. The app weighs how much of your formula's aromatic mass sits on measured values rather than estimates, and prints that next to the number: 40% when nothing is measured, up to 95% when everything is. It never prints 100, because this is still a model of a blend on skin. Tap the figure and it names the evidence.

What it cannot see. Skin chemistry, weather, how much you apply, and what the materials do to each other. Those are real and the model does not model them, which is most of the reason the ceiling is 95 and not higher.

And you can check it. After a wearing you can record what actually happened: how long it really lasted, and whether each phase came in weaker or stronger than predicted. The app then shows its own error on your own formulas. Nothing is adjusted automatically and none of it is uploaded; it is your record of how well the model does on your formulas, which is the only version of that question worth answering.

The percentages are not typed in by hand, and that is the point. The catalog stores the order a material's families come in (which one leads, which is secondary), and one shared curve turns that order into shares: the primary family takes 6 parts, the second 3, the third 2, the rest 1 each, with a family's share split across its sub-categories. That same curve drives the profile chart, the pyramid, the color of the material's swatch and the genre title, so all four agree because they are one calculation rather than four that happen to be maintained together. Inside a formula each material's share is then weighted by its dilution-adjusted aromatic mass, so a line weighed from a 10% solution counts for a tenth of the same gram taken neat.

One thing on a row is a job rather than a smell. 5 materials are marked as UV filters, and the reason is citrus. The terpenes carrying a cologne's whole opening are the first thing light takes apart: the color drifts, the top note goes flat, and a hand-made eau that was beautiful in March is disappointing by May. A perfumer building one should be able to find those four without already knowing to search for "octinoxate", so the catalog says it beside the material instead of leaving it in a paragraph. It is not a facet, because a facet is an odor word and this is not an odor, and it is not inferred from the text either: a text match would also catch a material described as photostable, which resists light rather than absorbing it, and those are opposite claims.

Where the constituent tables matter. An estimated breakdown of what a natural is made of is not trivia: it is how a limit written against a single molecule gets enforced on the oil carrying it, and how the same restricted molecule arriving from three different naturals is added up instead of being missed three times.

Two honest limits. Every regulatory join (IFRA, GHS, allergens) is by CAS number, and a material can supply one through its constituents as well as its own row: of the 569 with no CAS of their own, 244 carry no constituent table either, and those are the ones that return "nothing known" rather than "nothing applies". The app does not dress that up as a clean bill. And there is no physical chemistry on a catalog row: no molecular weight, no vapor pressure, no logP, no boiling point. The behavior in time comes from measured blotter readings and family, not from physical constants. Both are open work rather than solved problems.

What is deliberately not here. There is no multi-user team workspace with roles and permissions: an account belongs to one perfumer and syncs across their own devices. There is no live supplier catalog integration: supplier and price are fields you fill or import, not a feed, and nothing reorders for you. There is no accounting, invoicing or shipping. If you need a factory ERP, buy a factory ERP. PerfuMate covers formulation, compliance and the batch, which is the part an ERP usually does worst.

And the regulatory limits, which matter more. The IFRA data is Category 4 only (fine fragrance on skin), and the app puts every formula in that category: a lotion, a soap, a candle or anything for children sits in a different category with different limits, and this app will not compute them. It is fixed at the 51st Amendment and ships inside the app, so a new Amendment arrives as an app release. The GHS hazard set is indicative and unverified, reaching 398 of 2,211 materials. The safety data sheet is a draft that says so on its own face and needs sign-off by somebody competent in chemical safety classification; its flash point and transport class are estimated from the carrier solvent, not measured, so nothing should be shipped on them. There is no REACH, no CPNP or SCPN notification, no UFI or poison-center submission, no Product Information File and no US state disclosure. None of this replaces a regulatory consultant, and the app never says it does.

What the production chain costs. Make mode itself is free, as are the stock ledger, batches, the printable batch record, the EU allergen declaration and the INCI line, so a free account can weigh a batch, deduct what it used and print a record of it. Signing in, which is free, opens the rest: the safety data sheet draft, PDF export, compare and Make mode's smart weighing order. Two things need a paid plan, and they are separately gated: bottle labels, and the sale-ready pack that contains them. Otherwise the free limit is the formula count: three formulas at a time (the plans are here).

None of these are AI. They are deterministic engines with tests behind them, which is why they can be trusted with a number. AI Studio sits beside them and is optional.

Two things are worth being exact about. The backup is the whole history, not a snapshot: every saved version of every formula is in the ZIP, so a branch you abandoned in March is still there. And it is re-importable rather than just readable: importing backup.json restores all your data, which is what makes it a backup instead of a printout.

One thing deliberately does not travel: the passcode on a locked formula. A four or six digit code sitting in a file you email to yourself is not a lock, so a restored backup asks you to set it again. Everything else, including your private notes, costs and suppliers, is in the file.

It works in reverse too. PerfuMate imports formulas from PDF, CSV, TSV, JSON or text you simply paste, matches the material names against 5,888 supplier synonyms so "Phenylethyl alcohol" finds the row you already have, and takes several files or a ZIP of them at once. Whatever you can get out of your current tool, it will most likely read.

Monthly, Yearly and the AI Studio plans renew until you cancel. Canceling stops the next charge, not access you already paid for. On an AI plan, credits refill every 7 days, and unused weekly credits do not roll over.

Nothing is ever locked away from you: your formulas stay readable and exportable at every tier, including free (see what you can take with you). What a paid plan buys is unlimited formulas and exports. Every tool opens with a free account, with exactly four exceptions, all on a paid plan: bottle labels, the sale-ready pack that contains them, color PDF themes, and formula version history. The prices are here.

One thing gets assumed about PerfuMate that is not true, and it pushes people away from what would suit them. The free tier is not a trial: it does not expire, it does not ask for a card, and it does not need an account at all. Open PerfuMate and it is there. On price, Unlimited can be rented by the month or the year, or bought once and kept.

It installs to a phone straight from the browser: open PerfuMate, then Add to Home Screen. On Android it is also on Google Play.

Any phone can also install it from the browser. Open PerfuMate on the phone, then Add to Home Screen. On Android that is the browser menu; on an iPhone it is the Share button. It lands as an icon on the home screen and opens like any other app, without a store account, an install prompt or a download.

Neither way is a cut-down mobile version. It is the same app, with the same catalog, the same arithmetic and the same compliance checking, and if you sign in it holds the same formulas you left open on a larger screen. Weighing happens beside the scale, so the phone was the first screen this was designed for rather than an afterthought.

What happens if it stops. This is the right question to ask a small product, so here is the mechanical answer rather than a promise. The app runs on your own device: until you sign in, every formula is in a local database and the app works with no account and no network at all. Everything exports at every tier including free, in open formats, in one button: CSV any spreadsheet opens and JSON any text editor reads, with every saved version of every formula in the file. Nothing is in a proprietary format, nothing needs our server to be readable, and no export of your data is behind a payment; only the presentation extras are. If this site went dark tomorrow, a backup taken today would still open. The full list of what comes out.

It is worth knowing before you buy, in both directions. It means features asked for by people who actually make perfume tend to ship in days rather than quarters. It also means there is no support desk and no sales team, and a reply to a message comes from the person who wrote the code.

There are other things called Perfumate, and none of them are this one. There has been a perfume-finder website and an unrelated Android app of a similar name. PerfuMate is at perfumate.app and nowhere else: if a page, a listing or a download calling itself Perfumate is not on this domain, it is a different product and nothing to do with us.

What is not claimed: there is no company registration to point you at, no funding, no team, and no independent audit of the compliance data beyond the check described above, which you can re-run yourself from the repository. Treat the IFRA figures as a working set to compose against and verify against the official standards before you sell, exactly as the app stamps on every compliance sheet.