Swoonberry is the software for running your own dating service, city-wide or for a niche such as hikers or book lovers. Members aged 18 and over sign up on the phone or the web, pass an age gate, build a profile with prompts and photos, find people in a rules-based deck, match when both like each other, and chat only inside a match. Your team runs everything from a Trust & Safety console. It is built by DEV3 Technologies on Laravel 12 and Filament 4, with one Flutter app (the date app) whose source is included.
What it does
- 18+ only. An age gate asks for a date of birth before anything else, on the phone and on the server. Under-18 sign-ups are closed and the device is flagged; a ban is final.
- Profiles and prompts. Display name, age, optional pronouns, intents, interests, three prompts and two to six photos. Every photo goes through a moderation queue and shows as "In review" until a moderator approves it. There are no questions about orientation, religion or ethnicity.
- Discovery with distance bands. Filters by age, distance band, intents and interests. Location is stored only as a rounded, fuzzed cell and shown as a band such as "< 2 km"; exact positions are never shown. Ranking is plain rules (recent activity, profile completeness, verification, shared interests, boosts), not AI.
- Mutual-like matching. Two members who like each other at the same moment get exactly one match. Likes you is blurred until Premium, super-likes and a rewind are included, and a blocked pair never matches again.
- Chat inside a match. Messages, icebreakers, delivery receipts and typing on Laravel Reverb with a polling fallback. A scam and contact-detail filter holds suspicious messages behind a warning, and every chat photo is moderated before the other person sees it.
- Safety centre. Block, report, unmatch, safety tips, a manual pose-selfie verification reviewed by your moderators, and a date check-in timer that alerts a trusted contact.
- Trust & Safety console. Photo and verification queues with an SLA, reports with an audited context copy, flagged messages by reason code, a strike ladder, bans, device flags, appeals, blocked words, support tickets and API keys with moderator-bound scopes.
- Premium, boosts and super-likes. Subscription plans, boost and super-like packs on an idempotent credit ledger, proration on upgrade and pro-rata refunds. Purchases are simulated in the demo.
- Admin. Dashboard with queues and a neighbourhood chart, members with a timeline, plans, orders with refunds, subscriptions, the credit ledger, reports and funnel, safety content pages, settings, a key-based REST API with an OpenAPI document, and webhooks.
The Flutter app
The date app (apps/date) covers splash, phone sign-in, the 18+ age gate, the profile builder, location rationale, Discover cards with like, pass, super-like and rewind, filters, profile detail, Likes you, Matches, chat, the safety centre, premium and boosts, and a Me tab with pause, incognito, language (English, Spanish, French, German, Arabic), and account deletion. Location is coarse only (Android coarse permission, iOS reduced accuracy). It starts in demo mode and has a store-publishing guide in docs/STORE-PUBLISHING.md, including the dating-category and in-app purchase rules.
Gemini AI admin tools
Optional, bring your own key: add your own Gemini key after purchase (Google bills your account); with no key the AI buttons stay disabled. Every result is a draft you preview and edit; nothing is banned, removed, refunded or sent automatically.
- Reports and flags: summarise a report from its reason and redacted profile text, explain a flagged message from its reason code, suggest a decision for a human to make, and draft a reply to an appeal or ticket.
- Profile text queue: check a bio against your guidelines.
- Members: a timeline summary from event types and dates only.
- Plans and content: write plan and safety-page copy, suggest a price test from aggregate conversion, and suggest blocked-word variants.
- Dashboard and reports: explain the numbers, a report or an anomaly.
- A Google Translate language switcher in the public header, the member web app and the admin bar.
Message text is never sent to Gemini, photos and verification selfies are never sent, and names, emails, phone numbers and birth dates are replaced before any call. You remain responsible for your own privacy notices and your agreement with Google.
Quality checks
The release passed seven gates on the final code: Larastan level 6 with 0 errors, a Pest suite of 298 tests (1,926 assertions) that includes every Filament tab and AI action, a build with a small page budget and no 3D, migrations and the full demo seed on MySQL 8 and SQLite, a headless smoke test over 31 pages, 109 MySQL race assertions run three times (one match for a simultaneous like, reports decided at once, pack refunds, and more) plus real-websocket realtime checks, the web installer driven from the zip, and a secret scan. A consistency command (swoonberry:check --demo) proves matches, credits, entitlements and location cells agree on the demo data. The Flutter app passed analysis, 98 unit, flow and widget tests, and simulator builds, and ran once against the live API.
What you should know before buying
- 18+ only. The age gate is mandatory and under-18 sign-ups are closed. You remain responsible for dating-service, privacy and age laws where you operate.
- Location is shown as distance bands, never an exact position. Distance is approximate on purpose.
- Message text never goes to AI. Flagged messages are explained from the filter's reason code only.
- No saved notifications inbox in v1. Push and realtime events are delivered, but there is no stored list of past notifications.
- Every chat photo is moderated by a person before the other member sees it.
- Photo verification is a manual review of a pose selfie by your moderators; there is no face recognition or ID service. The scam filter is a heuristic, not a guarantee.
- Payments. Web card checkout through a gateway adapter comes later. Purchases are simulated in the demo, and the in-app purchase driver needs the buyer's store receipt verifier with
purchase_driver=iap. Keep the default simulated driver off any store build.
- No video dates and no AI matching or AI-written messages in v1. Demo profiles use illustrations, pets, landscapes and hobby objects, never real people.
- One Flutter app with source and a store-publishing guide. Building, publishing and store approval are your step; the app was verified in simulator builds, automated tests and against the live API, not on a physical device.
- Use MySQL or MariaDB for a live site; the concurrency checks were run on MySQL 8. One cron entry every minute is required.
Included
Source code, the web installer, the Flutter app source with a store-publishing guide, a buyer guide, demo data for a fictional service, Brambleside Social (about 6,000 members, 24,000 photos, 66,000 likes, 14,000 matches, 177,000 messages, 3,400 reports and 4,450 orders) and six months of support.