Verbatim doesn't run on a cloud server — the Whisper model runs on Akash's Mac, and this website reaches it through an encrypted tunnel. When the Mac is asleep or its connection blips, the engine briefly can't be located. It reconnects on its own; this page keeps checking every few seconds and will tell you the moment it's back.
Run these on the Mac to bounce the service and tunnel:
launchctl kickstart -k gui/$(id -u)/com.akash.transcribe
launchctl kickstart -k gui/$(id -u)/com.akash.transcribe-tunnel
Logs live at ~/Code/yt-transcribe/web/server.log and
tunnel.log next to it. Full documentation: ~/Code/yt-transcribe/README.md.
Install the toolchain and model:
brew install yt-dlp whisper-cpp
mkdir -p ~/Code/yt-transcribe/models && cd ~/Code/yt-transcribe/models
curl -LO https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo-q5_0.bin
Then re-clone ~/Code/yt-transcribe/web, reinstall the two
LaunchAgents (com.akash.transcribe, com.akash.transcribe-tunnel), and
restore /docker/traefik/dynamic/transcribe.yml on the VPS. The README in the project
folder documents the full architecture.