Yesterday's Top Launches: 5 Tools from September 23, 2026
Five products launched on September 23, 2026, including a decision model, a robosecretary, a creative testing platform, a Mac desktop organizer, and a cron scheduler, with two aimed at developers, one at marketing teams, and two at longstanding everyday problems.
Five products landed on September 23, 2026, and the spread says a lot about where things are headed: a decision model, a robosecretary, a creative testing platform, a Mac desktop organizer, and a cron scheduler. Two of them fit the new developer tools mould almost too neatly, one is aimed at marketing teams with big budgets, and two solve unglamorous problems that have annoyed people for years. Here’s what actually shipped.
Jev
TypeSafe AI’s Jev is the most conceptually interesting release of the day. The pitch: unstructured state goes in, typed probabilistic decisions come out. Instead of generating prose, it returns Choice, Score, or Noul answers with calibrated probabilities attached, which means your code can branch on the result without a parsing layer in between.
That last part is the real argument. Anyone who has wired an LLM into a decision path knows the awkward middle step where you coax a paragraph into JSON and hope the phrasing stays stable. Jev cuts it. The claimed numbers are striking too: roughly 70-500ms responses, 20-200x faster and 40-400x cheaper than comparable LLM workflows, with output tokens free. Whether those comparisons hold in your workload is another matter, and the “Noul” naming is going to confuse people no matter how well it works. Still, for teams running high-volume automated decisions, the shape of the thing is right. It’s free at console.typesafe.ai with no waitlist. See Jev.
Sai
Simular AI calls Sai the world’s first robosecretary, and it works through a fleet of autonomous computers that read the screen, click buttons, and type into forms the way a person would. The target is all that software nobody can automate: legacy desktop apps, internal portals, anything behind a login with no API in sight.
The approach isn’t novel in itself. Screen-level automation has been tried repeatedly, and the results have historically been fragile. What makes Sai worth watching is the 73% OSWorld score, which is a measurable signal rather than a marketing claim, plus Windows, macOS, and Linux support and a fleet model where one commander directs several machines. The honest caveat is that computer-use agents still fail in ways that are hard to predict, and a benchmark score won’t tell you how it handles your particular internal tool. But if you’ve spent years hearing that your systems “aren’t automatable,” this is aimed squarely at you. Free, via browser sign-in. See Sai.
AI Creative Insights by Decode
Entropik’s new platform tests ad creatives before media spend goes out, predicting attention, emotion, and conversion impact. Upload a banner, video, or OOH asset and it returns second-by-second emotional curves, visual hierarchy heatmaps, and prescriptive suggestions about what to change.
The synthetic audience feature is the most practical bit: build reusable personas like Gen Z Shopper or Value Buyer, then compare how the same creative scores across each one before committing budget. The reported outcomes are 95% predictive attention accuracy and 40% CTR improvement, which are vendor numbers and should be read as such. This is also the least accessible product of the five for a solo builder, since the entry point is a sign-up or a demo request with name, business email, phone, and LinkedIn. Marketing and research teams at larger brands will get the most out of it. See AI Creative Insights by Decode.
Turfs
Mac users have watched Windows users organize their desktops with Stardock Fences since 2009 and gotten nothing comparable. Turfs is the answer, and it’s a good one. It replaces Finder’s desktop icon layer with labeled areas called turfs, and each one can be backed by a real folder, a Finder tag, or a purely virtual collection.
The distinction between those three matters more than it sounds. Folder turfs physically move files, tag turfs show anything tagged from anywhere on the machine without moving it, and collection turfs are risk-free staging areas because nothing touches the disk. Native interactions stay intact, Quick Look and Cmd-I and Cmd-Delete all work, and the JSON-based local-first design means no account and no telemetry. The $29 one-time price with a 10-day trial is fair, though it does require macOS 14 Sonoma and it can never ship through the Mac App Store given how it hooks into the desktop. See Turfs.
Cronhq
Cron failures are silent by nature: a job stops running and nothing tells you. Cronhq targets exactly that, offering exactly-once execution enforced through Postgres locks so two workers can’t fire the same billing job twice.
The surrounding features are where it earns its keep. Webhooks carry HMAC-SHA256 signatures with rotatable per-job secrets, retries use configurable backoff, every execution is logged with status and response body, and alerts are deliberately quiet, firing on the third failure in an hour rather than the first, then once on recovery. Heartbeat monitors cover jobs Cronhq doesn’t run itself, which is a genuinely useful inversion. Built in Rust on Postgres, MIT-licensed, self-hostable, with a free tier of five jobs. The cronhq.yaml sync workflow is the part that will win over teams who want schedules in version control. See Cronhq.