Why I Built s1ft

English is not my first language. I’ve worked in tech for years, I think in English, I dream in English sometimes, but it’s not the language I grew up speaking. In professional settings, that gap shows up in ways you don’t always notice yourself.

I have a friend who’s honest with you after a meeting, not during it. He’d pull me aside and say things like: “Hey, you stressed the wrong syllable in ‘development’ again.” Or: “That phrase you used, that’s a direct translation from Russian. It doesn’t land the same way in English.”

He wasn’t being critical. He was being helpful. And every time he pointed something out, I got a little better. Not because I studied a rule, but because someone I trusted gave me specific feedback right after the moment it happened.

That kind of feedback was worth more than any course or textbook. Someone telling me what I actually said and how to say it better, while it was still fresh.

The problem with every tool I tried

At some point I realized that almost all of my professional communication happens in online meetings. Standups, syncs, client calls, 1:1s. Hours of me speaking English every day. If I could record and analyze those conversations, I’d have an endless stream of real-world practice material.

So I went looking. I found meeting tools: Fireflies, Granola, Summary AI, Voicenotes. They all do roughly the same thing. Most send a bot to join your call, upload your audio to their cloud, give you a transcript and a summary. $10–35 per month, whether you use them twice or two hundred times.

Three things bothered me.

The subscriptions. I don’t want to pay a flat monthly fee for something I use unevenly. Some weeks I have 20 meetings. Some weeks I have 3. I want to pay for what I actually use.

The bots and the cloud. A bot joins my call and everyone sees “AI Notetaker is recording this meeting.” The dynamic changes. People get guarded. And all that audio — my strategy discussions, my performance reviews, my client calls — sitting on someone else’s servers. I don’t want that.

No help getting better. This was the big one. These tools could tell me what I said. They couldn’t tell me how I said it, or how to say it better. I didn’t need another app that turns speech into text. I needed the thing my friend was doing, but for every meeting.

The lesson from childhood

I kept thinking about why my friend’s feedback worked so well. And I realized it followed the same pattern as the best learning I had as a kid.

The teachers who made the biggest difference weren’t the ones who lectured. They were the ones who watched me do something, then immediately told me what to fix. Try, hear what went wrong, try again. No delay. Direct correction in context.

That’s how people actually get better at things. You do the thing, someone tells you what to change, you do it again slightly better.

I wanted that for my professional communication. Something that watches every meeting with me and says: “You said ‘basically’ 9 times, try replacing it with a pause.” Or: “You started rushing after the Q&A.” Specific and timely, based on what actually happened.

Nothing like that existed. So I started building it.

Update, October 2026: s1ft no longer sends coaching emails or shows built-in coaching scores. It measures your voice on your Mac and lets you run your own prompts. The sections below were edited to match.

What s1ft does today

s1ft runs on your Mac. It captures system audio natively — no bot joins your call, no third-party participant in the attendee list. Transcription happens on-device using WhisperKit. Your audio files never leave your machine. And when you need to let participants know, built-in consent templates make it simple.

When you want a summary, s1ft sends the transcript directly from your Mac to the AI provider you choose: Anthropic or OpenAI using your own API key, or a local model through Ollama or LM Studio. There’s no s1ft server in the middle. If you never request a summary, your transcript text never leaves your Mac at all. s1ft also records the pitch and level of your own voice on your Mac. Agent skills work out pace and uptalk from those numbers; s1ft itself does not measure them. These are numbers, not scores or advice. s1ft doesn’t coach you on its own. If you want a review of how you spoke, you can install the review-my-delivery skill in your own AI agent and run it on a meeting. The coaching guide shows how.

Where this is going

The idea behind s1ft hasn’t changed: feedback built from your own meetings, not generic advice. Today that means measured voice data and agent skills where you decide what to ask. Whether and how s1ft goes further is still open.

I’m building this for non-native speakers, for professionals who want to sharpen how they come across, for anyone who’s walked out of a meeting thinking “I could have said that better” but had no way to know exactly how.

s1ft exists because the tool I needed didn’t exist.