On-device products launched on Friday, Sep 18, 2026
Products whose own page says the work happens on the device. Some of them are also cloud products; the page shows the overlap. Every number below was measured inside that day's window.
Friday in numbers
Friday, Sep 18, 2026- 546 products appeared across 7 public launch boards. We have read 546 of them end to end (100.0% of the day).
- Attention did not spread out: the busiest 55 products took 83.2% of the 5,290 votes the boards handed out, and the median product finished the day on 2.
- 11 of the 542 products on a board that publishes votes (2.0%) had not received a single one when we counted.
- 46.3% of the day put AI at the core of the product, while 30.0% never mention it at all.
- More of the day was aimed at consumers: 251 products said they sell to consumers and 40 said they sell to businesses, out of 437 that named a buyer at all.
- Nothing owns the words founders use for themselves: of the 519 products that named what they are, only 21 phrases were used by more than one of them, and the most common ("desktop app") by just 6. Against the 10 days before, "mac app" gained the most ground.
- 139 products printed a price. The middle one asks $9.99, and 10 of the printed price points were $0.
- 76 launch pages (13.9%) called the product the best, the first or the only one of its kind.
Every sentence above is assembled from a measurement made when this page was built. No number on this page is typed by hand, and no product is named: the report is counts only.
Was this a normal day?
Each measure against the 10 days before it: the bar is the range those days covered, the notch is their middle, the dot is this day.What kind of thing it is
A product can be more than one of these — a web app with a mobile client counts in both — so the shares add up past 100%.
- desktop34.4%188
- web34.1%186
- mobile23.4%128
- cli8.6%47
- extension5.9%32
- api2.2%12
- bot1.8%10
Read from each product's own pages, not from the board's category.
Who it sells to
One hundred squares are the whole day. The grey squares are products that said nothing about a buyer, and the palest are the ones we have not finished reading.
- consumers46.0%251
- developers17.4%95
- businesses and consumers9.3%51
- businesses7.3%40
- Read, said nothing20.0%109
Where it was published
App stores, extension stores and package registries named on the page.
- App Store15.9%87
- Play Store8.1%44
- npm3.7%20
- Chrome Web Store2.7%15
- GitHub1.5%8
- Microsoft Store1.5%8
- Homebrew0.4%2
- Mac App Store0.4%2
- PyPI0.4%2
- TestFlight0.4%2
175 products named a store or registry. The rest ship from their own site.
What the launch pages claimed
Three buckets, never two: says yes, says no, says nothing. Silence is not a no.
- Has a free plan63.4%
- Has documentation28.0%
- Is open source27.8%
- Puts a number on the page23.6%
- Asks you to contact sales14.1%
- Calls itself the best / the first13.9%
- Shows a case study2.2%
- Shows customer logos1.6%
- Says it is hiring1.3%
Every yes here is backed by a quote from the product's own page; a claim without one is not counted.
How the day's attention was shared out
5,290 votes across 542 products that appeared on a board which publishes votes.
The dashed line is what an even day would look like. The gap between the two is the day's inequality.
- top 10%83.2%4,399
- 10–20%5.5%289
- 20–30%2.4%128
- 30–40%2.0%108
- 40–50%2.0%107
- 50–60%1.0%54
- 60–70%1.0%54
- 70–80%1.0%54
- 80–90%1.0%54
- 90–100%0.8%43
Votes keep arriving after a product appears, so a fresh day is still filling up and an older one is closer to final. The measure is the sum of a product's votes across every vote-publishing board it was on that day, which is why the deciles add up to the day's total exactly.
What AI does in it
Whether AI is the product, a feature beside it, unmentioned, or explicitly ruled out.
- core to the product46.3%253
- not mentioned30.0%164
- a side feature13.7%75
- explicitly no AI1.1%6
- Read, said nothing8.8%48
"Never mentions it" is an absence of a claim, not a claim of absence. The two are counted separately here because conflating them once produced a headline that was wrong by two orders of magnitude.
Where it runs
Cloud, on the user's own machine, or self-hosted on the buyer's server.
- on the device, offline100.0%546
- cloud / hosted8.2%45
- self-hosted1.8%10
What size of company
Who the page says it is sized for. A page can name more than one.
- solo / indie11.5%63
- teams7.3%40
- small businesses4.9%27
- enterprise1.6%9
What the day charged
139 products printed at least one price. Bands hold the lower bound and exclude the upper: $10–29 means $10 up to but not including $30.
- $0100.0%8
- $1962.5%5
- $2.9962.5%5
- $4.99/month62.5%5
- $9/month62.5%5
- $9.9950.0%4
- $0.9937.5%3
- $19.9937.5%3
- $19/month37.5%3
- $29/month37.5%3
- $0/month25.0%2
- $1.9925.0%2
Prices in a currency other than the dollar are not converted; they are counted as unparsed and shown below. Making up an exchange rate on a page whose whole claim is measurement would be the most expensive kind of mistake here.
What the free tier is
Free forever, free up to a limit, or free for a while.
- free forever56.4%308
- free up to a limit4.8%26
- free for a trial4.0%22
- Read, said nothing34.8%190
What the trial asks for
What stands between a visitor and the product on the first click.
- nothing45.8%250
- an email address5.9%32
- a card0.7%4
- no trial0.7%4
Whether the page says we or I
Whether the launch page speaks as a company or as one person.
- “we”69.0%377
- “I”7.9%43
This is what the page calls itself, not what we know about the team: a one-person company can still write "we". It is a floor on how many are solo, never a count of them.
What founders called their own product
The short phrase each page uses for itself. Plurals and capitalisation are folded together; nothing else is. This is the closest thing we have to a search term, and it arrives earlier than one: a founder starts building when they see the need, a search happens once the product exists.
- 1desktop app1.1%+0.4 pts · ▲7
- 2tool1.1%-0.1 pts · ▲1
- 3mac app0.9%+0.7 pts · ▲25
- 4productivity0.7%+0.3 pts · ▲8
- 5browser extension0.5%-0.3 pts
- 6game0.5%+0.3 pts · ▲12
- 7menu bar app0.5%+0.2 pts · ▲8
- 8PDF tools0.5%+0.1 pts · ▲2
- 9utilities0.5%-0.5 pts · ▼5
- 10video editor0.5%+0.4 pts
- 11windows app0.5%new
- 12workbench0.5%new
- 13workspace0.5%+0.2 pts · ▲3
- 14app0.4%-0.9 pts · ▼12
- 3mac app0.9%+0.7 pts · ▲25
- 1desktop app1.1%+0.4 pts · ▲7
- 10video editor0.5%+0.4 pts
- 4productivity0.7%+0.3 pts · ▲8
- 6game0.5%+0.3 pts · ▲12
- 7menu bar app0.5%+0.2 pts · ▲8
- 13workspace0.5%+0.2 pts · ▲3
- 19desktop application0.4%+0.2 pts
519 products named what they are. 21 phrases were used by more than one of them, after folding plurals and capitalisation; everything else in the day was said once and is not listed.
Movement is measured as a share of each day's own cohort against the 10 days before, never as a raw count: as the corpus grows, every raw count rises and the page would read as if everything were trending. "New" means the term was below the threshold in those days, which is a warning and not a finding.
What the day leaned on
Products and platforms named on the launch page. Each one is also a dependency: if that API changes, this many pages have to change with it.
- Codex10.1%55
- GPT-6 Astra9.7%53
- GitHub9.0%49
- iPhone7.1%39
- macOS6.8%37
- ChatGPT6.4%35
- OpenAI6.2%34
- Windows6.2%34
- Claude4.8%26
- Google4.6%25
- Android4.4%24
- Claude Code4.0%22
- Chrome3.8%21
- Apple3.7%20
465 products named at least one other product or platform.
Who they said it was for
The audience named on the page, in the page's own words.
- 1developers5.7%-0.3 pts
- 2students1.8%-0.3 pts · ▲1
- 3freelancers1.3%-1.0 pts · ▼1
- 4designers1.1%+0.1 pts · ▲2
- 5creators0.9%-1.0 pts · ▼1
- 6founders0.9%+0.1 pts · ▲1
- 7parents0.7%+0.4 pts · ▲13
- 8professionals0.7%-0.4 pts · ▼3
- 9power users0.5%+0.3 pts · ▲16
- 10consultants0.4%-0.2 pts
- 11djs0.4%+0.2 pts
- 12marketers0.4%-0.4 pts · ▼3
- 13musicians0.4%no change · ▲2
- 14novelists0.4%+0.3 pts
139 products named what they are. 17 phrases were used by more than one of them, after folding plurals and capitalisation; everything else in the day was said once and is not listed.
Movement is measured as a share of each day's own cohort against the 10 days before, never as a raw count: as the corpus grows, every raw count rises and the page would read as if everything were trending. "New" means the term was below the threshold in those days, which is a warning and not a finding.
What the day was built with
Read from the live sites themselves, not from a survey: the analytics tag, the host's fingerprint, the payment script.
- GitHub Pages100.0%26
- Vercel69.2%18
- WordPress26.9%7
- Netlify11.5%3
55 sites showed one
- Google Analytics100.0%67
- Plausible4.5%3
- Umami3.0%2
71 sites showed one
- Gumroad100.0%5
6 sites showed one
Only what a page reveals about itself can be seen here, so this is a floor and not a census. A site behind a proxy or with no tag shows nothing and is not counted as "none".
Where the day was launched from
The place printed on the page — a company address, a legal footer, a founder's line.
- Singapore17.6%6
- India14.7%5
- United States14.7%5
- Australia11.8%4
- New Zealand8.8%3
- United Kingdom8.8%3
- Japan5.9%2
- Netherlands5.9%2
- Pakistan5.9%2
- Philippines5.9%2
34 of the 100 pages that printed a place named somewhere we could resolve to a country. The rest named a city we do not map, a region ("EU"), or nothing that resolves; they are excluded rather than guessed at.
Which boards the day came from
A product can appear on more than one board, so the column adds up past the day's total. Vote counts are read now, not at launch time.
| Board | Products | Share of day | Votes | Per product | Median |
|---|---|---|---|---|---|
| Product Hunt | 503 | 92.1% | 4,999 | 9.9 | 1 |
| PeerPush | 18 | 3.3% | 107 | 5.9 | 4 |
| Uneed | 12 | 2.2% | 26 | 2.2 | 1 |
| Fazier | 7 | 1.3% | 155 | 22.1 | 23 |
| BetaList | 4 | 0.7% | — | — | — |
| MicroLaunch | 1 | 0.2% | 2 | 2.0 | 2 |
| Tiny Startups | 1 | 0.2% | 1 | 1.0 | 1 |
Some boards publish a rank but no votes; those rows show a dash. An absent vote count is not a zero.
What it is about
The subject axis, assigned by a separate reading pass.
- Tools for working with AI coding agents9.3%51
- Mac and Windows desktop utilities8.1%44
- Tools for gamers7.9%43
- Tools for converting and compressing files4.0%22
- Quick developer utilities3.5%19
- Tools for notes and remembering3.3%18
- Tools that run AI locally3.1%17
- Tools for building AI agents2.7%15
- Tools for training and sport tracking2.7%15
- Tools for making video2.2%12
- Tools for habits and focus2.0%11
- Tools for courses and self-learning1.6%9
This axis is filled by its own pass, not by the main reading queue, so its coverage moves on its own schedule.
How many are still answering
2 days after the day closed, we asked every one of them again.
The denominator leaves out the ones we could not reach: a product behind a bot wall is not a dead product. Calling it dead would be the cheapest way to make this number look decisive and the fastest way to make it wrong.
How this issue was made
- The window
- One UTC day: Friday, Sep 18, 2026. A product belongs to the day its board says it launched, not the day our collector reached it — otherwise the report would print our own crawl speed as market news.
- The publishing gate
- A day is published only once the collector has run the morning after it, which is when that day's boards are read. Days without that run are not published and not quietly padded. 546 products cleared the minimum of 40.
- Where the answers come from
- Each product's own pages are read and answered question by question, with a quote required for every yes. 546 of 546 products were read all the way through before this issue was built.
- What is not here
- No product names, no rankings, no scores. A block whose cohort was read less than 70% of the way through is not printed at all, and says so where it would have been.