Private, enterprise-grade AI content-operations platform. Built for one operator (Stavarai), trained on real client data, positioned for acquisition.
This repo is internal. No client data, no internal architecture, no agent names should ever appear in public-facing surfaces (landing page, blog).
An AI-driven content engine for a Shopify-brand social media agency serving four niches: Food & Beverage, Beauty & Skincare, Apparel & Accessories, and Home & Lifestyle. It researches, generates, scores, and schedules social content at a volume no human team can match — then learns from what works.
The company never sees how it works. They see results.
No backend, no secrets, no setup. Renders the full UI with seeded data.
cd frontend
npm install
npm run dev
# open http://localhost:3000- Landing page at
/ - Blog at
/blog - Admin gate at
/admin— password:BLASTER2026
npm install # workspaces (frontend)
cd api && pip install -r requirements.txt && cd ..
npm run dev:local # FastAPI :8000 + Next.js :3000See docs/HANDOFF.md for the complete VPS setup, or run setup.sh on a
Linux server. Frontend deploys to Vercel.
frontend/ Next.js 15 + Tailwind v4 — landing, admin, blog
api/ FastAPI — auth, dashboard, clients, pipeline, settings, voice
rust_core/ AES-256 encryption, session auth, rate limiter, job queue
supabase/ Migrations — schema, client isolation, demo seed
skills/ grill-me, stop-slop, food-beverage, ugc-video, scoring
content/blog/ MDX posts (original, value-add)
agents/ Hermes SOUL.md + config
docs/ Specs, secrets contract, TDD spec, acquisition playbook
tests/ TDD — Rust + Python
.beads/ Reversible checkpoints (every destructive op logs one)
docs/CLAUDE.md— root orientation (ICM structure, 10 core laws)docs/PROJECT.md— who this is for, what it does, acquisition framingdocs/SECRETS.md— every env var, where it lives, never hardcodeddocs/TDD_SPEC.md— every feature as a test, before codedocs/HANDOFF.md— what runs now, what needs a VPS, the 7 gaps to fill
- Tests before code. Every feature has a failing test first.
- No client data mixing. Each client gets an isolated Supabase schema.
- Stop-slop on all generated text. No "AI-powered", no "revolutionize".
- LLM-agnostic. No hardcoded model names. Swap providers in settings.
- One bead per destructive op.
.beads/is the audit trail. - Rust for the hot path. Auth, encryption, rate-limiting, job-queuing.
BLASTER2026is the only backend access. Backend greets Stavarai by name.
- All API keys encrypted at rest (AES-256-GCM via Rust core)
- Supabase RLS on every public table
- Per-client schema isolation (verified by test)
- Telegram bot ignores every user except Stavarai's ID
.envis gitignored and chmod 600 on the VPS
Proprietary. All rights reserved. See docs/BUILT_TO_SELL.md.