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.
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
Needs Bun 1.4.3 (the pinned version; see packageManager), 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 :3000Clerk 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.
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 # APICI (.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.
Both apps run on Vercel: the web app as project blog-ai
(https://blog-ai.vivancedata.com) and the FastAPI backend as project blog-ai-api on
Vercel's Python runtime (apps/api/vercel.json). Database on Neon, auth Clerk
(clerk.vivancedata.com), billing Stripe. Dockerfile.backend / railway.json and the
GHCR image deploy.yml publishes are an alternative container path, not what production
runs. An hourly workflow probes the site's /api/health and the API's /health. Details
in docs/DEPLOYMENT.md, env in docs/ENVIRONMENT.md,
schema in docs/DATABASE.md, HTTP API in docs/API.md.
AGPL-3.0. See LICENSE.