What kinds of files can I upload?

Video (mp4, mov, mkv, webm), audio (mp3, wav, m4a, flac), images (jpg, png, webp, heic), and documents (pdf, txt, md). You can also paste a YouTube, Vimeo, Twitter, or TikTok URL and we'll download it for you.

How accurate is the transcription?

We use OpenAI Whisper. Word-level accuracy on clear English audio is typically 95%+.

Is my content private?

Yes. Files are stored encrypted in your private bucket on Cloudflare R2. Only you can access them. We don't train models on your data.

What happens when I run out of credits?

Existing content stays accessible — search and chat keep working with what you've already processed. New uploads pause until you top up or upgrade. No data loss, no surprise fees.

Do credits roll over?

Subscription credits reset each billing cycle. Top-up pack credits never expire.

Can I cancel anytime?

Yes — month-to-month, cancel from your account at any time.

Not yet — coming on the Studio tier later this year.

How long does processing take?

A 10-minute video is typically ready in under 90 seconds. Pro and Studio tiers use a priority queue.

Guide · April 15, 2026 · 6 min read

How to search inside videos with AI (in 2026)

For years, “searching a video” meant scrubbing a timeline. Now you can ask a video a question in plain English and jump straight to the answer. Here’s how the modern stack works — and how to set it up for your own library in under three minutes.

The old workflow was broken

Manual transcript review takes 20+ minutes per hour of audio. Folder trees are useless once you’ve got more than 30 files. The result: most of what you record is recorded once and never seen again.

What AI changes

AI video search fuses two signals that used to live in separate tools:

Transcription — every spoken word, with word-level timestamps. State-of-the-art is OpenAI’s Whisper, which hits ~95% word accuracy on clear English.
Frame description — every visual scene gets described by a vision model. Now you can search for “the moment she pointed at the chart” even if no one said the word “chart.”

When you fuse these into a hybrid full-text + vector index, the result is something that feels like Google for your entire personal video archive.

The 5-step setup

Upload your media. Drag and drop, or paste a YouTube / Vimeo / Twitter URL.
Wait for processing. Roughly 90 seconds for a 10-minute video.
Search in plain English. “every laugh in the founder podcast.”
Chat with your library. Cross-reference dozens of files in one query.
Cut and share clips. One sentence in, ready clip out.

Queries this unlocks

“Find every moment Ramsay called something disgusting” — pulls timestamped matches across episodes.
“Summarize Amjad’s argument about why most SaaS won’t survive” — answers with cited timestamps.
“Cut every clip where the founder talked about pricing” — returns ready-to-share segments.

Try it on your own library

videan does all of the above out of the box — Whisper for transcription, a vision model for frame analysis, Claude Sonnet 4.6 for chat, and a hybrid retriever for search. You get 5,000 credits free, which is enough to upload one real video and run a full search → chat → clip flow end-to-end.