Built so customer conversations
don't get lost in calendars.

Why async voice feedback

Discovery calls are expensive. Two people block 30 minutes, someone reschedules, and a third of the booked slots ghost. Forms are cheap but lifeless. You get short text and no tone.

Voice carries the hesitation and the frustration that text strips out. Async voice keeps that signal and removes the calendar tax. People talk faster than they type, so you get longer answers from a wider sample.

How HeySpeak works

You create a Magic Link with one question. You share it by email, LinkedIn, or text. The recipient opens it in any browser, hits record, talks for a minute, and sends. There is no signup, no app, and no calendar.

On your side, every voice note is transcribed by Voxtral and summarized by Mistral. You scan the AI summary in seconds, open the transcript when you want detail, or play the audio when you want to hear the tone. The whole loop, from sending the link to reading the summary, takes minutes.

Who builds HeySpeak

HeySpeak is built by Thomas Lorenz, a product builder based in Bremen, Germany. Background in political science, seven years of work across no-code, AI tooling, and B2B SaaS. The team page has the long version.

Where the data lives

Every layer sits in the EU. Hosting on Vercel in Frankfurt, database and auth on Supabase in Zurich, audio in a private Cloudflare R2 bucket, transcription and summaries handled by Mistral AI in France. Voice data is never used to train AI models, never sold, and accessible only through 1-hour signed URLs scoped to the link owner.