Senja

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 testimonial system that replaces Senja: collect what customers say, keep the proof, and put it on my site without a third-party script.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- ffmpeg on the box for video, or nothing at all if you only take text
- Caddy in front, serving media directly from disk

THE DATA MODEL
- spaces: id, slug, name, brand_json, collect_page_json
- testimonials: id, space_id, kind, author_name, author_title, author_company, author_avatar_path, body, rating, video_path, poster_path, duration_ms, source, source_url, status, consent_text, consent_at, created_at
- imports: id, space_id, platform, external_id, raw_json, imported_at — the original of anything pulled in from elsewhere
- walls: id, space_id, slug, layout, filter_json, theme_json
- wall_views: id, wall_id, created_at, referrer_host, ip_hash — because a wall that nobody sees is worth knowing about
- Kinds: text, video, imported. Status: pending, approved, hidden
- consent_text is the exact sentence the person agreed to, stored with the row. A testimonial without recorded consent is a legal problem waiting

COLLECTING
- A public collect page per space: a question, a text box or a record button, and the fields you actually need
- Video recorded in the browser with MediaRecorder, uploaded in chunks, transcoded to H.264 and a poster frame taken at two seconds
- A hard limit on length — 120 seconds — enforced in the browser and again on the server
- The consent checkbox is not pre-ticked, and its exact wording is stored with the testimonial
- A thank-you screen with an optional offer, because the moment somebody has just praised you is the moment to ask for one more thing
- Works on a phone, in one hand, without an account

IMPORTING WHAT ALREADY EXISTS
- Paste a URL from a review site, a social post or a support ticket and store it with a screenshot of the source
- The screenshot is the proof: a deleted tweet still needs evidence that it was said
- Imported testimonials are marked as such and carry a link to the source. Never present an import as if it were collected

APPROVAL
- Nothing is public until somebody approves it
- Edit for length and typos, never for meaning, and keep the original body alongside the edited one so the change is visible
- Tag by product, by theme, by customer size
- A rejected testimonial is hidden, not deleted

PUTTING IT ON A SITE
- A wall: a public page and an embeddable version, both server-rendered
- Layouts: grid, masonry, single quote, carousel that is a scroller and not a timer
- The embed is an iframe with a fixed set of layouts, or a script that injects plain HTML. Both take a height message so the host page can size the frame
- No web fonts, no framework, under 15KB
- Every wall filters by tag, rating or kind, and the filter is in the wall's definition rather than in the embed code
- Structured data for reviews on the wall page, so a search engine can read it — and only where the testimonials are genuine and consented, which they are, because that is how they were collected

MEDIA HANDLING
- Store original and transcoded, serve the transcoded one
- Avatars resized to two sizes at upload
- Everything on your own disk with hashed filenames; nothing on a CDN you do not control

THE INTERFACE
- One screen: the queue on the left, the testimonial being read on the right, approve and tag from the keyboard
- Dark and light, one accent

OPERATIONS
- .env: DATABASE_PATH, MEDIA_PATH, BASE_URL, SMTP_URL, ADMIN_PASSWORD_HASH
- Migrations on boot, each once
- Backup covers the database and the media directory, and the restore script proves both
- Health endpoint that checks disk space, because video fills a disk quietly

WHAT MATTERS MOST
Collection and consent are the product. Build the collect page, the consent record and the approval queue first, and record a real video on a real phone before writing a line of the wall. A testimonial you cannot prove was given freely is one you cannot use.

Give me the repository, migrations, .env.example, a seed space with six testimonials, and a README with deploy steps behind Caddy including ffmpeg.