Only 7 of 345 described launches say where the maker is based

Only 7 of 345 described launches say where the maker is based

The launch that taught us most about our own board is Cronhq, a hosted scheduler for HTTP calls that appeared on 21 September 2026 at version 0.1, with a free plan and a paid tier from $5 a month. Its site claimed an MIT licence and self-hosting, and pointed at a code repository. When we checked on 22 September, that repository returned a 404 and the package named in the install instructions did not exist on the public registry. The site named no company and carried no terms or privacy policy. Everything a buyer would normally lean on to judge a $5 tool was either missing or pointed at nothing.

Cronhq is an extreme case, but it sits at the end of a pattern we can now count. Our launch board holds 444 published launches. After setting aside the 19 that come from Rutba, a software company whose launches are on the board, 425 remain. Of the 345 that carry a written description, 7 state where the maker is based. Of 37 products that call themselves open source, 29 name no licence at all. Of 253 product launches, 72 state a price in currency. And one in five rests on the maker's own word alone, a share that climbs to 30 of 36 for products found through Hacker News.

This article sets out what we counted, how, and where the figures should be read with care. It is about the board as it stands on 9 October 2026, not about the software market at large.

How we counted 425 launches

Every launch on the board has a type: product (253 once Rutba is set aside), AI model (170) or major update of an existing product (2). Launch dates run from 11 June to 8 October 2026, with the heaviest months being August (117) and September (207). We wrote each description ourselves from the maker's own pages and any press we could find, and where the evidence is thin the description ends with a red note beginning "Not yet independently verified:", which says what is missing.

Every figure below comes from a script, article-method.js, that reads the published records and applies fixed patterns: a currency amount for price, the words "open source" or "open weights" or a named licence for licensing, and phrases such as "based in" or "registered in" for location. Red notes are matched the same way. The patterns are blunt, so three cautions apply.

First, the counts describe our descriptions, which in turn describe what the maker's pages said when we read them. A maker may state its location somewhere we did not look. Second, the pattern for location catches explicit statements and will miss a looser phrase such as "a Berlin team", so the 7 is a floor, not an exact tally. Third, we compare shares across small groups, and a group of 36 moves by almost three points with every launch.

Definitions used here. A launch "states a price" if its description gives a currency amount. "Open wording" means the description says open source, open weights, self-hostable or names a standard licence. A "red note" is our standard note that independent confirmation is missing. "Described" means the entry has a written description at all.

Not every entry has one. 80 of the 425 have no description, and 65 of those are among the 67 launches dated before 1 August, loaded in July with a name, a tagline and a link and little else. Since 1 August, only 15 of 358 lack a description. That matters for every share below, so each is worked out on described entries only, and we say so in each place.

Where the maker is based: seven of 345

Location is the thinnest fact on the board. Among 235 described products, 6 give a place for the maker, and among 108 described AI models, 1 does. That is 2.0 percent of described entries.

The places that do appear read like a world map drawn by accident. siift is based in Vancouver, VoiceCap in Vilnius, Kennel in Munich and Peas and Pans in Dehradun. Agnes 3.0 Flash comes from a Singapore-based team, and Privatrak is registered in Wyoming. Snagr, covered below, gives only a country. Where a place appears, the maker put it on its own pages and we repeated it.

The red notes tell the other half of the story. Of the 87 launches with one, 22 say that the company, operator or location is not given. Snagr, a tool that chases abandoned checkouts, is run by an individual in India according to its own terms, and no company is behind it. Emote, an API that lets chat agents react with an emoji, is made by Experimental Software, which is not described as an incorporated company and gives no address or governing law. Verirun, a command line tool, publishes no company, terms or privacy policy and sits on a subdomain of another product's domain. My App Engineer publishes terms that choose Dutch law but names no operating company.

Why does this matter for a buyer? A business that signs a processing agreement needs a legal entity on the other side, and a security questionnaire asks where data is held. A launch that cannot answer "who and where" in a sentence has not yet reached the point where a procurement team can even start. The failure is quiet. Nothing on these pages looks wrong. The page just stops short of the question.

Open source, with the licence left out

Open wording is common. Of 235 described products, 48 (20.4 percent) use it, and among 108 described AI models the share is 49.1 percent, which is 53 entries counting self-hosting language. The sharper number is what follows the claim. Of the 37 products that say open source in so many words, only 8 name a licence from the MIT, Apache, GPL, AGPL, BSD or MPL families. The other 29 leave it open.

That gap has a practical cost. Open source is a category of licences, not a single permission, and the difference between MIT and AGPL decides whether a company can embed the code in a paid product. Bruno, the API client that stores collections as plain files, and Hedwig, an MIT-licensed cold email tool, are both described as open, but only one of those descriptions carries a licence name that lets a lawyer answer the embedding question without a click.

The models behave better. Among the 40 described models that say open weights or open source, 24 name a licence, and across the whole board Apache 2.0 appears in 27 descriptions and MIT in 20. Model makers publish a model card with a licence field as a matter of routine. Product makers often have only a repository page, and the licence file may be there while the sentence on the launch post is not. We also noticed that 39 of 170 models carry a link to downloadable weights, which is a stronger signal than the word "open" in a description.

One more cross-cut: 13 of the 48 products with open wording also state a price. Open and priced are not opposites. Several run a hosted service on top of free code, and the price is for the hosting.

What prices look like when they are stated

Of 235 described products, 72 (30.6 percent) state a currency amount. Counting the 18 products with no description, the share of all 253 is 28.5 percent. AI models do better, at 54 of 108 (50.0 percent), because the model makers publish a price per million tokens as part of the release: 49 of the 108 described models give one.

For products, 47 give a dollar price per month. They range from $4 for Inbound, an email API, to $599 for the largest of the credit plans that Firecrawl's Alexandria runs on. The middle of the 47 is $29. Seventeen sit at or under $20, 25 between $21 and $100, and only 5 above $100. The tool a launch board sees is overwhelmingly a small, self-serve subscription, not a sales-led enterprise contract.

Two oddities stand out. Twenty-six of those 47 prices end in a 9, and 18 are exactly $19, $29, $49 or $99. Retail psychology has reached software pricing so thoroughly that a price like $13.99 a month, which is what Jottoo charges for its one plan, now looks almost exotic. Ten products describe one-off or lifetime pricing instead of a subscription; AskDeck sells a finished presentation for $29 once, with no subscription.

The reverse of a stated price is not a hidden price. In 31 described products the text says a tier exists only by quote, as an enterprise plan or through a sales contact, or that no price is published. A buyer can read those as a signal about who the product is for. Snagr and Cronhq both publish a free tier, and 87 of 235 products mention free access of some kind (37.0 percent), a larger group than those that publish an amount.

The launches that rest on the maker's own word

Of 425 launches, 87 (20.5 percent) carry a red note. The share is nearly identical for products (20.9 percent) and for AI models (20.0 percent), which surprised us, because models arrive with announcements from large companies and products arrive with a post from a founder. Among entries with a description, it is 22.6 percent for products and 31.5 percent for models. The share rises because the two groups have different amounts of undescribed entries.

What is typically missing is easy to read in the notes:

• No press report, independent evaluation or coverage found: 64 of 87 notes (73.6 percent).
• The launch date rests on a post or listing the maker submitted: 51 (58.6 percent).
• The company, operator or location is not given: 22 (25.3 percent).
• Benchmark figures, scores or an evaluation that is the maker's own: 19 (21.8 percent).

The strongest split is by where the launch reached us. Of 36 products whose source link is a Hacker News thread, 30 carry a red note (83.3 percent). Of 114 products sourced from Product Hunt, 17 do (14.9 percent). Of the 22 products sourced to the vendor's own page or press, 5 do (22.7 percent). That is not a verdict on the forum. A Show HN post is by definition the maker introducing the product, and in all 30 of those notes the date rests on the maker's own post. If a product is dated by a Show HN post and nothing else, our rules say we say so. The honest reading is that Hacker News is where we meet makers early, often within days of a first public build, before anyone else has written about them.

The same early stage explains Hedwig, the MIT-licensed cold email tool introduced in a Show HN post on 30 September 2026. Our note says the site names no operating company or location and no independent report was found. A buyer could use it today, but nobody outside the maker has yet said that it works.

Red-noted entries are also the ones we researched hardest, which produces a result we did not expect. Of 53 red-noted products, 33 state a price. Of 182 without one, 39 do. The thinly sourced launches are the best priced on the board, because we went to the pricing page precisely when nothing else could be found.

Tinfield 1, a model from Bad Theory Labs, shows what the note is for. Its score on a public coding benchmark was self-reported, it was absent from the official leaderboard on 22 September, and the lab's own site described it as "not built" that day before presenting it as available without a release date on 1 October. A model can be dated by its repository timestamps and still not be something a team can use.

Product launches and model launches read like different genres

Product and model entries are different in what they bother to say. Of described entries, 75.0 percent of models mention a benchmark or score against 7.7 percent of products. Models mention a context window in 38.0 percent of cases, products in 0.9 percent. The median description before the red note runs 210 words for a model and 112 for a product.

Products talk about the buyer: 57.0 percent say who the product is for, against 9.3 percent of models. Products mention an API in 21.7 percent of cases, models in 56.5 percent. The Model Context Protocol, the connector standard that lets an agent call outside tools, appears in 42 product descriptions and in 2 model descriptions.

Of the 253 products, 116 (45.8 percent) mention an agent in the tagline or description, and 110 (43.5 percent) sit in developer tooling by our keyword count. Media, voice and video tools follow at 54 (21.3 percent), analytics and data at 50 (19.8 percent), and meeting notes and documents at 49 (19.4 percent). Security and compliance is 37 (14.6 percent). The categories overlap, since a developer tool for agents counts in both.

Does the agent label change what a maker tells us? Barely, on price: 31.0 percent of agent products state an amount, against 30.3 percent of the rest. On verification there is a gap. Among agent products 25.0 percent carry a red note, against 20.2 percent of the others. It is a small group and a small difference, and we would not build a rule on it. We include it because it runs the way a sceptic would expect.

Jottoo, which turns recorded meetings into deadline-tracked tasks with no bot joining the call, shows the typical product entry well: a stated price, a stated trial, a stated data position, and a note that it is a 2026 product on its first Product Hunt launch. Lev, an open-weight decision model from Interfaze, shows the typical model entry: a licence, benchmark numbers that are the maker's own, and a red note for exactly that reason.

What the exclusion of Rutba changes

Rutba is a software company whose launches are on the board. Its 19 entries (16 products and 3 major updates) are excluded from every count above, and we recomputed each headline share both ways.

• Launches with a red note: 87 of 444 (19.6 percent) with, 87 of 425 (20.5 percent) without. Rutba has none.
• Launches stating a currency amount: 126 of 444 (28.4 percent) with, 126 of 425 (29.6 percent) without.
• Launches with open wording: 103 of 444 (23.2 percent) with, 102 of 425 (24.0 percent) without.
• Launches without a description: 80 of 444 (18.0 percent) with, 80 of 425 (18.8 percent) without.
• Product launches: 269 of 444 (60.6 percent) with, 253 of 425 (59.5 percent) without.
• Source link on Product Hunt: 116 of 444 (26.1 percent) with, 116 of 425 (27.3 percent) without. On Hacker News: 36 of 444 (8.1 percent) with, 36 of 425 (8.5 percent) without.

The one count that moves enough to mislead is major updates, which would read as 5 of 444 with Rutba and is 2 of 425 without. Three of the five major updates on the board are Rutba's, and 18 of its 19 entries link to its own domain as their source. Including them would make the board look better sourced and slightly less priced than it is, and we do not want our own entries to help the board look better.

Limits worth stating

Our patterns can mistake things. A description may mention "$5 million" as a customer figure and be counted as a price, though we checked the sample of monthly prices by name and found the 47 above real. The count of products that say "open source" depends on our wording as well as the maker's. The 36 Hacker News products are too few for a firm rate. And a board of launches is a sample of what is announced, not of what is sold: nothing here says how many software products exist or what share of them publish prices.

The finding that survives those cautions is the pattern across all of them. The facts a buyer needs to start a purchase, which are who runs it, where, under what licence, and for how much, are published for a minority of launches, and the minority differs by fact. Price is stated by about three in ten products. A licence name follows about one open-source claim in five. A location follows one description in fifty.

Points a buyer could not guess

• A repository link on a launch page is a claim, not evidence: Cronhq's pointed at a 404 and a package that did not exist, on a product priced from $5 a month.
• A product that says open source names its licence only 8 times in 37, so ask for the licence file before you plan to embed the code.
• The cheapest tools are the best documented on price: the middle of the 47 monthly prices we hold is $29, and 26 of them end in a 9.
• Launches with the thinnest outside coverage were the most likely to publish a price (33 of 53 against 39 of 182), so a visible price says nothing about maturity.
• 30 of 36 products introduced through Hacker News had only the maker's own post to date them, against 17 of 114 from Product Hunt.
• Models state benchmarks in three of four descriptions while products do so in under one in twelve, so a score comparison across the two groups compares different kinds of evidence.
• Removing the company's own launches changes the shares by about one point, and changes the count of major updates from 5 to 2.