What I use
The tools that actually ship the work — daily drivers, the production stack, and what I pay for. Updated when it changes.
Daily drivers
- Editor — VS Code, increasingly driven through Claude Code for anything bigger than a one-file change.
- Terminal — Windows Terminal with PowerShell 7; Git Bash for POSIX scripts.
- AI assistant — Claude (Max subscription). Code, architecture, writing, and the agent pipelines in my projects all route through it — the subscription-CLI trick is what makes 31-pipeline apps like ToCoach cost $0 marginal.
- Browser — Chrome.
- Version control — Git + GitHub; GitHub Actions for every deploy.
Production stack (what actually ships)
- Frontend — Next.js 14, React 18, TypeScript, Tailwind CSS. Electron when it needs to be a desktop app.
- Backend — FastAPI (Python) for AI-heavy services, Node.js/Express (TypeScript) for product APIs.
- AI layer — Claude API + Claude CLI, Ollama for local models, deterministic-first routing before any LLM call.
- Databases — PostgreSQL for anything relational, MongoDB Atlas for product data, Redis for queues/cache, SQLite for local-first apps.
- Voice AI — Whisper STT on-device (ONNX), Kokoro-82M TTS on-device.
- Infra — Docker + Compose, GitHub Actions CI/CD, Firebase Hosting, PM2 on VPS, AWS S3 for files.
Hardware
- Machine — Windows 11 laptop that runs Whisper, Kokoro, and Ollama locally without complaint.
- Audio — a decent USB mic matters more than people think when you build voice AI all day.
Services I pay for
- Claude Max — $100/mo — routes all LLM inference through the local CLI; zero marginal token cost across every project.
- GitHub — private repos + Actions minutes.
- Firebase — this site's hosting + Firestore (free tier does a lot).
- Domain — legendev.org.
- Brevo — transactional + outreach email (free tier, used hard by LeadForge).
/uses convention — see uses.tech for hundreds more.