blog-AI：用多阶段 LLM 流水线生成并发布品牌口吻长文

Generates long-form blog posts and books in a trained brand voice, then publishes them. Live at https://blog-ai.vivancedata.com. Next.js web app, FastAPI backend, browser extension.

The design decision that matters: generation is a pipeline of small LLM calls, not one prompt. A post is researched (web search, up to 8 cited sources), outlined, written one section at a time with the brand voice summary and research context, then proofread and humanized. Two opt-in stages follow: an SEO loop that rewrites until a score threshold is met, and a claim-by-claim fact check. Each stage is a plain function in apps/api/src/ with its own tests, so a bad stage is diagnosable and swappable. Providers (OpenAI, Anthropic, Gemini) sit behind one generate_text() with retries and per-operation rate limits.

Layout

apps/web/        Next.js 16 app (Clerk auth, Sentry, Neon via route handlers)
apps/api/        FastAPI: app/routes/ are handlers, src/ is the generation logic
apps/extension/  Browser extension
db/migrations/   SQL schema, applied by `bun run db:migrate`
scripts/         migrate, seed, runtime audit, post-deploy health check

Run locally

Needs Bun 1.3+, Python 3.12, and one LLM API key.

git clone https://github.com/gr8monk3ys/blog-AI.git && cd blog-AI
bun install
cp .env.example .env                       # set OPENAI_API_KEY (or another provider key)
cp .env.local.example apps/web/.env.local  # NEXT_PUBLIC_API_URL=http://localhost:8000

cd apps/api
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python server.py                           # API on :8000

cd ../.. && bun dev                        # web on :3000

Clerk is optional in development: with no NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY every route is public. Without DATABASE_URL, history, brand profiles, and analytics fall back to in-memory storage. See docs/ENVIRONMENT.md for the full list.

Test

bun run lint && bun run type-check && bun run test:run && bun run build   # web
bun run test:e2e                                                          # Playwright
cd apps/api && DEV_MODE=true OPENAI_API_KEY=test pytest -q                # API

CI (.github/workflows/ci.yml) runs all of the above plus per-route coverage floors, a fresh-install migration smoke test, and DB-backed tests against Postgres 16.

Deploy

Web on Vercel, API as a container (Dockerfile.backend, pushed to GHCR by .github/workflows/deploy.yml on green main), database on Neon, auth Clerk, billing Stripe. An hourly workflow checks the deployed site's health. Details in docs/DEPLOYMENT.md, schema in docs/DATABASE.md, HTTP API in docs/API.md.

License

AGPL-3.0. See LICENSE.