Honest opinions, no paid placements. Some links are affiliate — every affiliated card is badged, and the disclosure is below.
Model providers and SDKs I call from production code.
Primary LLM for writing, coding, and agentic workflows.
Claude 4 series is my default for long-context reasoning and code review. I call it from the Vercel AI SDK for the kpboards blog tooling and from Claude Code for day-to-day engineering.
Alternatives: OpenAI, Google Gemini
Fallback model provider for image generation and TTS.
I keep GPT-4o wired up as a secondary model for cost-sensitive batch jobs and use their image / Whisper endpoints when Claude is not the right tool.
Alternatives: Anthropic Claude, Replicate
Typed streaming + tool calling for React 19.
Every AI feature on kpboards goes through the AI SDK. `streamText` + `useChat` just work with React 19 Server Components, and the provider abstraction lets me swap Claude / GPT without touching callsites.
Alternatives: LangChain.js, Raw fetch
Voice generation for Remotion video voiceover.
My Remotion explainer pipeline calls ElevenLabs during render: one API request per scene, cached as MP3 next to the composition so re-renders are free.
Alternatives: OpenAI TTS, Azure Speech
Hosted open-source models when I need a specific weight.
I reach for Replicate when a closed API does not cover the use case — Flux for illustrations, Whisper variants for long-form transcription. Pay-per-second billing keeps experiments cheap.
Alternatives: Hugging Face, Fal.ai
How I render programmatic video for explainers and YouTube content.
Render video with React. The core of every kpboards explainer.
Remotion is the one unfair tool in my stack. Every scene is a React component — version-controlled, type-safe, diffable. Drag-and-drop editors cannot touch this workflow.
Alternatives: Motion Canvas, After Effects
Fast rough-cut editor for screen recordings.
When a teardown or walkthrough needs real footage, I rough-cut it in Descript — transcript-based editing cuts an hour of timeline work down to fifteen minutes. Final polish still happens in Remotion.
Alternatives: CapCut, ScreenStudio
Auto-captioning for vertical shorts.
I use Submagic exclusively for captioning short-form output. Its word-level timing and emoji density beats every open-source ASR + SRT pipeline I have tried.
Alternatives: OpenCut, CapCut
Where every kpboards project actually runs.
Where kpboards.com runs. Zero-config Next.js 16.
The site is deployed on Vercel. The value is not the compute — it is preview URLs on every PR plus first-party support for React 19 Server Components and the App Router before anyone else.
Alternatives: Netlify, Cloudflare Pages
Postgres + Auth + Storage in one managed bundle.
kpboards stores every article, category, and lead in Supabase. I use the JS client directly from Server Components for reads and keep a service-role key for admin writes — no ORM overhead, full Postgres when I need it.
Alternatives: Neon, Planetscale
S3-compatible object storage with zero egress fees.
Rendered MP4s from the Remotion pipeline land in R2. Cheap storage, free egress, and a signed-URL flow that slots into Next.js image/video loaders without any adapter.
Alternatives: AWS S3, Backblaze B2
Transactional email that does not hate developers.
Newsletter confirmations, contact form replies, and lead-magnet drops all go through Resend. React Email templates render server-side, the API key is one env var, and their domain auth UI took five minutes.
Alternatives: Postmark, SendGrid
Container hosting for long-running Remotion renders.
Remotion renders that exceed Vercel function limits run on a small Railway worker. One-click Dockerfile deploy, predictable pricing, and logs that stream to my terminal.
Alternatives: Fly.io, Render
Editors, agents, and trackers I use every day.
Agentic CLI I use as my pair programmer.
The kpboards repo is driven heavily through Claude Code — agents for code review, SEO auditing, and i18n validation live in `.claude/`. It is where I spend more hours than any editor.
Alternatives: Cursor, Aider
Editor with first-class AI completions.
For quick edits, one-off scripts, or when I just want tab-completion backed by a long-context model, Cursor beats VS Code + Copilot. I keep both installed and pick based on the task.
Alternatives: VS Code + Copilot, Zed
Solo-founder issue tracker that feels instant.
Every content bucket, monetization plan, and bug lives in Linear. The keyboard-first UX matters when you are the only person using the tool — Jira slows one-person teams down.
Alternatives: GitHub Projects, Height
JavaScript runtime + package manager, faster than npm.
kpboards runs on Bun for install + dev + test. `bun install` is three-to-four times faster than npm on this repo and the built-in test runner handles Vitest-compatible specs without extra config.
Alternatives: pnpm, npm
Design surface + the storefronts I ship digital products through.
Where layouts start before they become Tailwind.
I prototype every new landing page in Figma first. One file per section, variants for mobile/desktop, then I translate to Tailwind by eye — faster than iterating in code.
Alternatives: Penpot, Sketch
Static landing page host for one-off launches.
When a product needs a standalone launch page that I do not want to maintain inside the Next.js monorepo, Framer ships it in a day with custom code components for anything interactive.
Alternatives: Webflow, Vercel
Merchant of record for my digital products.
The Remotion Starter Pack and Next.js AI template sell through Lemon Squeezy. Tax is handled globally, licences are issued automatically, and the checkout embed drops into any page.
Alternatives: Gumroad, Paddle
Backup storefront for community drops and free downloads.
I host free lead magnets and community-priced drops on Gumroad. Cheaper fees for tiny ticket sizes, and the ratings social proof carries well inside indie-hacker audiences.
Alternatives: Lemon Squeezy, Stripe Payment Links