Five Launches From October 12: Voice Pipelines, AI Visibility, and a Notch Full of Buddies

Yesterday’s batch of new developer tools skewed toward infrastructure rather than surface-level apps. Three of the five sit underneath existing workflows — a production layer for AI voice, a scoring model for agent decisions, and a browser-based capture tool that folds logging into the recording itself. The other two are more personal: a one-man GEO platform and a Mac app whose entire interface is a pair of earbuds. Here’s what shipped.

Onepin

The pitch here is refreshingly narrow. Onepin isn’t another text-to-speech engine; it’s the layer that sits on top of the thirty-odd engines you might already be paying for — ElevenLabs, OpenAI, Google, Azure, Cartesia, Murf and a long tail beyond that — and fixes the boring stuff that ruins otherwise good audio.

The problem it targets is real if you’ve ever shipped voiceover in more than one language. Models read “St.” as Saint when you meant Street, “3042” comes out as three thousand forty-two, and Siobhán turns into something nobody recognises. Onepin normalizes the script before synthesis, checks names against a four-million-word pronunciation dictionary (AbbVie, Måneskin, semaglutide, esophagogastroduodenoscopy), then scores every generated line on naturalness, word accuracy, clarity and pronunciation. Lines that miss get repaired word by word instead of forcing a full re-render.

Teams doing product videos, dubbing or course content across five languages will feel the difference immediately. Everyone else can probably wait. Billing is pooled across workspaces with free seats, which is a sensible touch for agencies juggling several clients.

ReSO AI

Generative engine optimization has been threatening to become a category for two years, and ReSO is one of the more complete attempts at it. The premise: buyers now ask ChatGPT and Perplexity for vendor recommendations, and most brands have no idea whether they appear in those answers.

ReSO splits the work into three buckets — an intent model that tracks how LLMs interpret buyer questions, citation intelligence that shapes content around what AI platforms actually cite, and a technical engine that audits whether your site is readable by these systems in the first place. The case studies are specific: SMC Global at 132% more ChatGPT clicks, an HRTech platform going from 10 to 50 AI mentions in three months.

Worth flagging that the numbers all come from the company’s own reporting, and the “replace your SEO agency” framing is aspirational at best. Still, the free $500 visibility audit makes it easy to test the diagnosis without committing.

Microsoft-Decision-1

Quietly the most interesting launch of the day. Microsoft released a decision-scoring model that does one thing: given a fixed set of options, it returns a calibrated probability for each in a single pass, as structured output.

That’s a deliberate break from general-purpose LLMs. Routing, classification, prioritisation and workflow control all need something an agent loop can act on without parsing prose, and they need it fast. Microsoft claims the fastest measured median latency — 2.5x quicker than the runner-up and 35x quicker than GPT-6 Sol — plus top accuracy across 36 benchmarks covering roughly 147,000 questions. Calibration came in third, about 1.5 behind the leader.

The model is post-trained from Qwen3.5-9B and available through Microsoft Foundry and OpenRouter at $0.042 per million input tokens, with output free. If you’re building agent infrastructure, that pricing alone justifies a look.

Cue

A free macOS app that treats your headphones as the focus switch. AirPods in, Focus mode turns on, distracting sites get blocked, timer starts. AirPods out, break begins.

It’s the kind of small idea that either fits your habits perfectly or does nothing for you. The execution is charming — everything lives in the notch, including a now-playing widget, per-app volume and a small army of animated buddies that roam your screen when you’re off the clock and hop back into the notch when work resumes. There’s a floating 190×32 pill for Macs without a camera cutout.

Underneath the whimsy sits a genuinely useful privacy stance: no accounts, no analytics, data stays local, written in Swift. The caveats matter though. It’s un-notarized, so first launch requires a manual approval in System Settings, and Face Unlock is explicitly convenience-grade rather than secure. Requires macOS 14 on Apple Silicon.

Skreno

Another browser-based screen recorder, but the differentiator is what travels with the link. When you stop recording, Skreno captures the network and console logs behind whatever broke, attaches failed requests with their error bodies, and hands you a shareable URL with a transcript, AI summary and second-accurate comments.

That combination targets a specific and genuinely painful loop: a bug report that arrives without a stack trace, followed by three replies trying to reproduce it. The Studio editor runs in-browser with up to 4K exports, one editor handles trimming, captions, blurring and backgrounds, and edits update the same link rather than requiring a re-upload. Free tier includes the full editor, no watermarks anywhere, and no account to try it.

The tradeoff is trust — feeding console logs into a shared link is convenient until you remember what’s sometimes in them.

Quick links

• Onepin
• ReSO AI
• Microsoft-Decision-1
• Cue
• Skreno