Scribly turns long videos into accurate, timestamped transcripts, works out who is speaking, and lets you search every sentence — then export to SRT, DOCX or PDF, or share a link that plays from the exact moment.
Self-hosted and private — the media never leaves the box it runs on.
What it does
Built for the files that are too long for the usual tools: podcasts, streams, lectures, interviews.
Word-level timing that stays in sync from minute one to hour four — no drift towards the end of long files.
Voices are separated and labelled automatically, then you rename them once and every segment follows.
Full-text search across every segment, including partial words. Click a hit and the player jumps to that second.
SRT and VTT for editors, DOCX and PDF for the people who will never open an editor.
Send a link and anyone can read along, seek the audio and download the file — no account needed.
Runs on your own server with quantised models on consumer GPUs. No upload to somebody else's API.
How it works
MP4, MOV, MKV, MP3, WAV, FLAC — the format is detected from the bytes, not the extension.
Long files are processed in chunks on the GPU so memory stays flat, then speakers are separated across the whole recording.
Search for a phrase, click to hear it, rename the speakers, download the format you need.
Questions
Hours, not minutes. Three-hour-plus recordings work today; chunked processing keeps memory flat so nothing falls over mid-file.
Anything ffmpeg can open, up to the per-plan size limit. Corrupt or half-downloaded files are rejected up front rather than after an hour of processing.
Onto the server running Scribly — a local dataset with snapshots and an off-box copy. Nothing is uploaded to a third-party transcription service.
Well enough for real interviews and panel streams. Speaker separation is automatic, and you can rename speakers yourself at any time.
The plan is a limited free tier for short videos plus a paid tier for long files, diarisation and bulk upload. Pricing lands with the public launch.
Yes — that is how it is built. A Proxmox host, a Postgres instance, a Redis queue and a GPU (or patience on CPU) is the whole stack.
You can use the demo instance today — upload a file and see the transcript, speaker labels and share page for yourself.
Open the demo