A second brain for video & audio
Point Cerul at your folders, YouTube channels, and podcasts. It indexes everything you watch and hear — on your own machine — so you search by meaning and jump straight to the moment.
The global overlay — one keystroke, search your library from anywhere
Most of what you learn is locked inside video and audio.
Talks, podcasts, recorded calls, tutorials — the hardest content to search. A transcript captures what was said; the rest is locked inside files you'll never scrub through again. Cerul turns them into searchable, local-first memory.
Find the moment, not just the file
Search by meaning
Hybrid retrieval — full-text (SQLite/FTS) plus vectors — finds the moment by meaning, not just keywords. Every result highlights the matched line.
Local-first
Media, transcripts, and the vector index all stay on your disk. Inference runs through your own keys, or fully local models — no Cerul account required.
Always on, out of the way
A global hotkey overlay, menu-bar tray, background indexing, and start-at-login. It stays one keystroke away.
⌥ SpaceMany sources, one index
Local folders, YouTube channels and videos, and podcast RSS feeds — point Cerul at them and it watches and indexes locally.
Three steps to searchable memory
Point it at sources
Add a folder, a YouTube channel, or a podcast feed. Cerul watches them locally.
It indexes locally
Extract audio, transcribe (your key or a local model), write the full-text index, then embed.
Search the moment
Search by meaning in the overlay, jump to the exact timestamp, play it inline.
A local API your agents can call
Cerul Core exposes a local REST API on 127.0.0.1:7777. Coding agents, scripts, and automations query your own library directly — without anything leaving your machine. Open-source, FSL-1.1-ALv2.
# search your own library by meaning
curl -X POST 127.0.0.1:7777/search \
-d '{"q": "what they said about scaling laws"}'
# → { "hits": [{ "item": "...", "t": "20:23" }] }One memory layer for video. Two ways to use it.
Local, open, yours
- Indexes and searches on your own machine
- Desktop app + global overlay + local API
- Free · self-hosted · your keys
A hosted video-search API
- Video search the way agents call it
- Managed indexing, account-backed sync
- Built for teams and scale
Before you download
Does my data leave my machine?
Media, transcripts, and the vector index all stay on your local disk. Transcription/embedding runs through your own provider keys or local models. The only exception is visual understanding (slides/charts) — an opt-in beta via Gemini, off by default.
Do I need an account? Is it free?
No account needed; the app is free and open-source. Transcription/embedding uses your own provider keys (you pay the provider directly), or fully free, offline local models on Apple Silicon.
Which platforms?
macOS (Apple Silicon) today, with signed & notarized installers. Windows and Linux are coming soon.
Does it work offline?
Yes — choose local models and it runs fully offline on Apple Silicon. With remote provider keys, only the transcription/embedding step needs the network.
How does it relate to cerul.ai?
Cerul App is the local, self-hosted desktop app (this site). cerul.ai is the hosted video-search API for AI agents. The app works fully standalone; a Cloud account is optional.
Your machine, your data.
Media, transcripts, and the vector index stay on disk. Visual understanding (slides, charts) is an opt-in beta via Gemini. No telemetry on what you watch.
Turn everything you watch and hear into searchable memory.
Download free, bring your own keys or run local models. Find your first moment in minutes.
macOS (Apple Silicon) today · Windows & Linux coming soon