English · 中文
Distribution amplifier for solo AI creators.
Submit your AI project once. Our agents auto-write 10+ platform-native posts in ~10 seconds — X, Reddit, Hacker News, Dev.to, LinkedIn, Bluesky, Threads, Xiaohongshu (小红书), Jike (即刻), and more. Edit inline, one-click publish, track engagement across all channels from one dashboard.
You vibe-coded an AI app. Building was the easy part.
Then comes distribution. Same product, 10 different platform conventions: X wants ≤ 270 chars, Reddit wants self-aware tone, Hacker News wants technical depth, Xiaohongshu wants emoji-led aesthetic notes, LinkedIn wants narrative, Dev.to wants long-form. Writing all 10 properly is 3-4 hours of context-switching that happens right when you're exhausted from shipping.
Solo founders skip platforms. The best one becomes the only one. Reach collapses.
Submit URL → Claude writes 10+ platform-native drafts (~10s)
→ Review inline, edit, approve
→ 1-click open platform with text pre-filled
→ Mark posted, paste the URL
→ Cron scrapes engagement every 6h
→ Dashboard shows which channel converts best
Bilingual by default. EN and ZH drafts generate from one submit — Western indie hackers + Chinese AI creators (小红书 / 即刻 / 知乎 / B站) reached without rewriting.
Platform-aware prompts. Each draft is generated with platform-specific system prompts: hook rules, length constraints, anti-marketing-fluff voice, no repeated numbers, language-output directive. Tuned via dogfood eval on real launches.
Real engagement tracking. Public-API scrapers for Reddit / HN / Dev.to / Bluesky / X. Cron writes append-only snapshots so you see the 30-day curve, not just current totals.
Live: vibexforge.com — sign in with GitHub or Google, paste any URL, get 10+ platform-native drafts in ~10 seconds.
Runs in 30 seconds. Zero config. No API keys, no database, no sign-up.
git clone https://github.com/alex-jb/vibex.git
cd vibex
npm install
npm run devOpen http://localhost:3000. The app runs in mock mode with built-in demo data.
Want real data? Wire up Supabase + Claude (optional)
cp .env.local.example .env.local
# Fill in:
# NEXT_PUBLIC_SUPABASE_URL
# NEXT_PUBLIC_SUPABASE_ANON_KEY
# ANTHROPIC_API_KEY # Claude Sonnet 4.6 for drafts
# OPENAI_API_KEY # gpt-image-2 for Xiaohongshu covers
# BLOB_READ_WRITE_TOKEN # Vercel Blob for cover storage
# CRON_SECRET # Bearer token for cron routes
# SUPABASE_SERVICE_ROLE_KEY # Cron writes
npm run devRun the SQL migrations in .private/migrations/*.sql through the Supabase Dashboard SQL editor. Full setup in CONTRIBUTING.md.
| Layer | Choice | Why |
|---|---|---|
| Framework | Next.js 16 (App Router, Turbopack) | RSC + streaming + edge runtime where it helps |
| Language | TypeScript 5 (strict) | Refactor fearlessly |
| UI | React 19 + Tailwind v4 + Framer Motion | Zero-config, fast, smooth |
| DB | Supabase (Postgres + RLS + Realtime) | One stack, one auth, realtime drafts |
| Auth | Supabase Auth (GitHub + Google OAuth + PKCE) | Cookies, not localStorage |
| Drafts | Claude Sonnet 4.6 + prompt caching | Parallel generation, ~10s for 10+ drafts |
| Translation | Claude Sonnet 4.6 (daily batch) | EN ↔ ZH for creator-submitted content |
| Visual | OpenAI gpt-image-2 | On-demand Xiaohongshu cover generation |
| Storage | Vercel Blob | Cover images + demo videos |
| Resend + Vercel Cron | Welcome / drafts-ready / weekly digest / daily summary | |
| Engagement | Public scrapers (Reddit, HN, Dev.to, Bluesky, X) | Cron every 6h, append-only history |
| Monitoring | Sentry + HyperDX | Server errors + browser session replay |
| Analytics | OpenPanel | Cookieless product analytics |
| Deploy | Vercel + GitHub Actions | git push = prod |
vibex/
├── app/ # Next.js 16 App Router
│ ├── page.tsx # Landing — IH-style, 10+ drafts pitch
│ ├── arcade/ # Original RPG splash (preserved)
│ ├── how-it-works/ # Marketing landing for inbound traffic
│ ├── dashboard/ # Creator's projects + drafts pipeline
│ ├── launch/ # Submit a project → 10+ drafts in ~10s
│ ├── project/[id]/
│ │ ├── drafts/ # HITL review UI: edit / re-roll / publish
│ │ └── analytics/ # Per-platform engagement + 30D sparklines
│ ├── ideas/ # Idea Lab (Claude-scored pre-build)
│ ├── admin/metrics/ # Launch-day monitoring (drafts + funnel)
│ └── api/
│ ├── projects/[id]/generate-drafts # Triggers parallel Claude calls
│ ├── drafts/[id]/reroll # Single-draft regen (cost-gated)
│ ├── drafts/[id]/generate-cover # Xiaohongshu cover via gpt-image-2
│ └── cron/
│ ├── scrape-engagement # Every 6h: refresh views/likes/comments
│ ├── translate-zh # Daily: EN → ZH for ideas + projects
│ ├── daily-owner-summary # Daily: traction email per creator
│ └── weekly-digest # Weekly: rollup email
├── lib/
│ ├── draft-generator.ts # Platform prompts + Anthropic client
│ ├── engagement-scrapers.ts # Reddit/HN/Dev.to/Bluesky/X public APIs
│ ├── visual-generator.ts # gpt-image-2 + Vercel Blob upload
│ ├── translate-zh.ts # Claude bilingual translator
│ ├── i18n.ts # ~925 strings + Lang context
│ └── i18n-categories.ts # Render-time category localization
├── proxy.ts # Supabase SSR auth middleware
├── public/llms.txt # AI search engine discoverability
└── scripts/dogfood-vibex-launch.ts # Eval harness for prompt quality
VibeXForge is the hosted product. These are the open-source Claude Code skills that came out of the same agent stack — they install as /<command> in any Claude Code session and work on any machine.
| Skill | Install | What it does |
|---|---|---|
| council-diff | npm install council-diff |
5 voices (Garry / Naval / PG / Suster / Cuban) answer a decision in parallel; Fable 5 Oracle adjudicates; Brier-score per voice over 30 days. The disagree-and-resolve shape for product decisions. |
| polymarket-brier | clawhub install polymarket-brier |
Forecast a Polymarket question with Haiku, persist the prediction, Brier-score after resolution. Per-source calibration table tells you whether the loudest forecaster was actually right. |
| solo-founder-os / splunk_obs | pip install solo-founder-os |
Splunk HEC adapter for cron-scheduled AI agents. Three sourcetypes (reflection / eval drift / bandit regret), 6-panel dashboard.xml ready to import. |
All three are MIT, all three have GitHub Releases, all three Brier-audit their own quality over time. The pattern is the same: build the thing, persist the predictions, score against reality, publish the calibration table.
Shipped
- Submit project → Claude writes 10+ platform-native drafts in ~10s
- HITL review UI: inline edit, single-draft re-roll, status filter
- 1-click publish — text auto-copied + intent URL opened per platform
- Engagement tracking — 6h cron scrapes Reddit / HN / Dev.to / Bluesky / X
- Cross-platform analytics with 30-day sparklines + winning-channel call-out
- Bilingual EN ↔ ZH (UI + auto-translated creator content via daily cron)
- Daily Claude cost gate (per-creator quota, atomic Postgres RPC)
- Xiaohongshu cover image generation (gpt-image-2 + Vercel Blob)
- PWA installable (manifest + minimal SW, no HTML cache)
- Email retention loop (welcome + drafts-ready + daily summary + weekly digest)
- /admin/metrics launch-day dashboard
Roadmap
- Stripe Connect payouts (85/15 take rate)
- Per-platform visual generation v2 (X OG cards, Bilibili thumbnails)
- Real OAuth auto-publish (X / Reddit / Bluesky API)
- Multi-account support (creator's personal X + brand X)
- Push notifications for new engagement milestones
PRs welcome. See CONTRIBUTING.md for setup, branch / commit conventions, and the dev loop.
Found a security issue? See SECURITY.md — please don't file a public issue.
Source Available — see LICENSE. Free to read, fork, and self-host. Commercial redistribution requires permission — DM @alex-jb.
Alex (@alex-jb) — solo indie AI creator. VibeXForge exists because I shipped my own AI gallery for 6 months and got 5 users; distribution killed me. So I built the tool I wished existed.
If it works for you, tell me (DMs open). If it doesn't, tell me louder.