About Figscribe

Why it exists

Voice notes are great for the person sending them and a bit of a chore for the person receiving them. Figscribe is a small, no-login tool that turns an audio clip into a readable transcript and a short summary - so you can skim a five-minute ramble in seconds.

It is intentionally a single-purpose tool. No sign-up. No database. No history kept on our servers. If that's all you need, it should feel quick and out of the way.

How it's built

Figscribe is a Next.js 16 app deployed on Vercel. Audio uploads go straight to Vercel Blob from the browser, get streamed through Groq for transcription (Whisper-large-v3-turbo) and summarisation (Llama 3.3 70B), then the blob is deleted. The transcript and summary are streamed back to the browser line by line.

The source code lives on GitHub - it's a small Next.js app with a single API route, a same-origin proxy lockdown, Upstash Redis for rate limiting, and Sentry for crash reports. There is no analytics beyond cookieless page-view counting.

Who made it

Figscribe is built by Ollie Brennan, a software engineer based in the UK. It's a personal project, hosted on the free tiers of Vercel and Groq, with no commercial intent for now.

Limits and quirks

• Audio files up to 25 MB (Whisper's per-request limit).
• 5 transcriptions per hour per IP, to keep costs sane.
• English by default; switch language in the picker for better accuracy on other languages.
• No background processing - the page must stay open while a transcript is running.

Feedback

Suggestions or bug reports are welcome at olliedoes.dev.