VideoAsk

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me a video form that replaces VideoAsk: I ask on camera, people answer on camera.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- MediaRecorder in the browser for capture, ffmpeg on a worker for transcoding
- Object storage with a delivery network in front
- whisper.cpp locally for transcription
- Caddy in front

THE DATA MODEL
- forms: id, slug, title, steps_json, theme_json, status
- steps: within the form — id, kind, prompt_video_id, prompt_text, answer_kind, is_required, logic_json — answer kinds are video, audio, text, choice, file
- responses: id, form_id, token, started_at, submitted_at, contact_json, ip_hash
- answers: id, response_id, step_id, kind, media_id, text, duration_ms, answered_at
- media: id, path, sha256, kind, duration_ms, width, height, poster_path, transcript_path, status
- Every answer stored as its own clip, because that is what makes them watchable in a list

RECORDING IN A BROWSER, WHICH IS THE HARD HALF
- getUserMedia for camera and microphone, MediaRecorder into chunks written out as they arrive
- **Upload chunks during the recording**, not after. A person who records ninety seconds and then waits ninety more for an upload abandons
- A crash loses seconds rather than everything
- **A live audio level meter, and a warning if it has been silent.** A silent recording is discovered at the worst moment, and it is the most common failure
- A countdown, a visible indicator, and the ability to re-record before submitting. People always re-record the first one
- Permission denied is the common case: say what to do about it rather than showing a broken player
- Codec support differs by browser. Transcode on the server and accept whatever arrives

ANSWERING WITHOUT FRICTION
- No account, ever. A token in the URL
- **An option to answer by text instead**, always. Not everybody can record video — a shared office, a bad connection, a disability, or simply not wanting to be on camera. A form that only accepts video excludes people, and the ones it excludes are not random
- Works on a phone in portrait, which is where most answers come from

TRANSCODING AND DELIVERY
- A queue with a concurrency cap; this is CPU-heavy and unbounded parallelism takes the machine down
- One rendition and a poster frame is enough for a short clip; a full ladder is for long video
- **Signed expiring URLs.** These are recordings of people's faces answering questions, and a guessable path is not acceptable
- Storage is the real cost — measure bytes per minute and set a retention per form

TRANSCRIPTION, WHICH IS WHAT MAKES ANSWERS USABLE
- Every answer transcribed locally, shown beside the clip and searchable
- **Reading forty answers is far faster than watching them**, and without transcription a video form produces a pile nobody reviews
- Export the transcripts as CSV alongside the media

PRIVACY
- Say clearly, before recording, what happens to the video and how long it is kept
- A retention with a sweeper that actually deletes, and deletion on request by response token
- No third-party transcription without saying so

WHAT MATTERS MOST
Chunked upload during recording, and a text alternative on every step. The first is why people finish; the second is why everybody can.