Orbit Social AI Content Operation

A production-minded, multi-tenant content operations platform that turns a brand brief into reviewed, scheduled, and published social content. The first release defaults to mock publishing and strict AI budgets so it is safe and inexpensive to run.

What works

• Monorepo with Next.js web, NestJS API, Prisma/PostgreSQL, Redis/BullMQ-ready infrastructure, and shared packages.
• Brand profile → structured AI variants → edit/review → schedule → idempotent mock publish → analytics placeholder.
• Eight connector adapters: Facebook, Instagram, LinkedIn, TikTok, YouTube, X, Threads, Pinterest.
• Tenant-scoped data model, Supabase-compatible JWT verification, RBAC schema, encrypted OAuth token vault, rate limiting, audit model, health checks, retries/idempotency primitives.
• AI request hashing, caching contract, structured outputs, usage metering, workspace budget cap, output cap, and inexpensive model routing.

Quick start

1. Copy missing values from .env.example into the private .env. Never commit .env.
2. Start infrastructure: docker compose up -d postgres redis
3. Install: pnpm install
4. Generate Prisma: pnpm db:generate
5. Apply the development schema: pnpm db:push
6. Run: pnpm dev
7. Open http://localhost:3000; API health is at http://localhost:3001/v1/health.

Optional n8n: docker compose --profile automation up -d n8n.

Validation

Run pnpm check. End-to-end: pnpm --filter @sma/web exec playwright install chromium, then pnpm --filter @sma/web test:e2e.

Safety defaults

• Social publishing is mock unless SOCIAL_CONNECTOR_MODE=live.
• Production requires valid Supabase JWTs and an x-workspace-id.
• OAuth tokens use AES-256-GCM; use a secret manager for TOKEN_ENCRYPTION_KEY.
• Duplicate AI prompts resolve through the request hash cache contract.
• The default workspace AI cap is USD 10/month and each response is capped.

See docs/ for architecture, integrations, security, deployment, costs, and operations. Project links and decisions are maintained in WIKI.md.