Uses
The tools I build with
The hardware, apps, AI and open-source repos behind my mobile apps, my AI agents and the content system that runs my business. Only what I actually use, and what each one is for.
Updated November 2026
The desk
MacBook Pro 16-inch
Apple M3 Max · 14-core CPU · 30-core GPU · 96 GB
My main machine. Agents, builds, video renders and the local speech models all run on it, and it is the edit bay of the Affarit podcast studio.
Mac mini
The second Mac, so agents can run in parallel while I work on the first.
MacBook Air
The light one, for when the 16-inch stays on the desk.
iPhone
Every build of my phone apps lands on it first, and its Apple Health data feeds the life OS I run my business from.
The podcast studio
Sony FX3 + 24-70mm GM II · 4 × Shure SM7B · RØDECaster Pro II
Camera, mics and console of the studio I built at Affarit, with three-point LED light and acoustic treatment.
See the studioWhere the work starts
Almost everything I build now starts as a sentence to an agent. These are the agents, the models and the formats that keep them on a leash.
The agent I work with all day, in the terminal and inside VS Code: my apps, both websites, my business notes and my content system.
More about Claude Code Read: The AI SDLC I Use to Ship Mobile FeaturesThe models behind my apps. Every call goes through one metered table: Opus writes, Sonnet does most of the work, Haiku sorts and triages.
More about it Read: Meter Every AI Call From Day OneRuns several Claude Code agents side by side on my Mac, each in its own copy of the repo.
More about Conductor Read: Three Agents at Once Locked My App's AI Out for Four HoursFolders of instructions an agent loads only when the task needs them. My setup is full of them: video, design, writing, documents.
More about Agent SkillsHow my agents reach tools and data outside the repo, one narrow tool at a time.
More about it Read: MCP for Mobile Developers
What my agents can reach
The MCP servers my agents talk to, two of them my own. Each one hands an agent tools, never the keys: read-only where it can be, one token per run where it can't.
Claude Desktop and the coaches in my app read my notes through it, each coach only the tools its role allows. It has no write tool, and health and family notes are never served.
More about My life OS, read-only Read: MCP for Mobile Developers: Give Your Agent Tools, Not AccessThe Copilot in my content system's video editor edits through the editor's own tools, served as an MCP server for one run, with its own token for that run.
More about itFigma's remote server: Claude Code reads a design's frames, components and variables instead of guessing from a screenshot.
More about Figma MCP serverClaude works in my own Chrome, in tabs of its own: it reads pages and clicks through them.
More about Claude in ChromeConnects the agent to my editor: it sees the open file and the editor's errors and warnings.
More about itGoogle Drive for documents, and Booking.com, Expedia and Kiwi.com when I plan a trip.
More about Claude connectors
Building and shipping
The editor, the stack under this site and affarit.com, and the two desktop apps I wrote for myself.
My editor, with Claude Code running inside it.
More about Visual Studio CodeEvery project lives in git, every change an agent makes is a commit I can read, and the gh CLI and GitHub Desktop do the rest.
More about GitHub Read: My AI Agents Can't Write. They Propose, I Accept.This site and affarit.com: Next.js 16, React 19 and Tailwind v4, with every page and article written as code.
More about itPostgres for both sites: sign-ups, leads and the email log.
More about SupabaseHosts both sites. A push to main is a deploy.
More about VercelEvery email both sites send, and every send logged.
More about ResendMy two desktop apps: the life OS I run my business from, and the content system that makes my posts.
More about ElectronRenders this site's article covers in a real browser.
More about Playwright
Native Android and iOS
Mobile is my trade: more than ten years of Android and iOS apps, native first and cross-platform when it fits.
Native Android.
More about itNative iOS.
More about Swift and SwiftUI Read: Six Crashes Importing Apple HealthCross-platform when it fits. The phone app of my own life OS runs on Expo.
More about React Native and Expo Read: Hermes Has No JIT: Why Your React Native App Scrolls but Taps HangAndroid builds, the emulator and the profiler.
More about Android StudioiOS builds, signing and Instruments.
More about XcodeA regular in the Android and iOS apps on my CV.
More about FirebaseApple's icon library, for anything that ships on iOS.
More about SF SymbolsMoves files to and from an Android phone on a Mac.
More about OpenMTP
Video, made in code
My reels and explainers are written, drawn and rendered by code on my Mac, and the voices and transcripts run locally.
Reels as React components: every scene is code, rendered on my Mac.
More about Remotion Read: How I'm Building an AI Marketing FactoryHTML in, video out, made for agents to write.
More about HyperFramesFilm styles made in code for Claude Code: watercolour, ukiyo-e, data stories and many more.
More about Lemo-OpuscarTranscripts with word timings, on the Mac, with Whisper large-v3-turbo.
More about whisper.cppA small, open text-to-speech model that runs locally.
More about KokoroFast local text-to-speech.
More about PiperCuts, mixes and contact sheets, under everything else.
More about FFmpegPulls a video down so I can study how it was made.
More about yt-dlpFor the edits I still do by hand.
More about Final Cut Pro
Notes, calendar and the rest
One person, a few businesses: the system is plain files and agents that propose.
Clients, tasks, money and decisions live in a folder of markdown files. Agents read it and propose; I accept.
More about it Read: I'm Moving My Whole Business Onto Markdown and Claude CodeSynced both ways into that folder, so an agent sees my week the way I do.
More about itMeeting notes, transcribed on my Mac.
More about MeetilyWhere Engineering OS, my newsletter, is read.
More about SubstackMy eBook and the Digital Toolkit.
More about Gumroad
A curated list of repos
The open-source projects my stack stands on, and the ones I saved and kept after reading the README. Each one checked on GitHub in November 2026.
In my stack
- anthropics/claude-code
Anthropic's agentic coding tool. It runs in the terminal and inside editors like VS Code, reads a codebase, edits files and runs commands, within the permissions you give it.
- anthropics/skills
Anthropic's format for packaging instructions, scripts and files into a folder that an agent loads only when a task calls for it. The repository holds the examples and the document skills.
- vercel/next.js
Next.js is Vercel's React framework for websites and web apps, with server rendering and static pages. React is the UI library under it, and Tailwind CSS a utility-first way to style it.
MIT - supabase/supabase
An open-source backend built on Postgres: database, auth, storage and row-level security, hosted or self-hosted.
Apache-2.0 - electron/electron
A framework for building desktop apps for Mac, Windows and Linux with JavaScript, HTML and CSS, on Chromium and Node.js.
MIT - microsoft/playwright
Microsoft's framework for testing and automating web pages in Chromium, Firefox and WebKit.
Apache-2.0 - expo/expo
React Native builds native Android and iOS apps with React. Expo is the framework and tooling around it: native modules, builds and updates.
MIT - ganeshrvel/openmtp
A free, open-source Mac app for moving files to and from Android devices.
MIT - remotion-dev/remotion
A framework for making videos programmatically with React: every frame is a component, rendered to a video file.
- heygen-com/hyperframes
An open-source tool from HeyGen that renders HTML compositions to video, built for agents to write.
Apache-2.0 - lemomo-ai/lemo-opuscar
A toolbox of film styles made in code for Claude Code, ready to use or to extend with your own.
- ggml-org/whisper.cpp
A C/C++ port of OpenAI's Whisper speech recognition model, fast on Apple silicon, with word-level timestamps.
MIT - hexgrad/kokoro
Kokoro-82M, an open-weight text-to-speech model with 82 million parameters, small enough to run on a laptop.
Apache-2.0 - OHF-Voice/piper1-gpl
A fast, local neural text-to-speech engine, now developed by the Open Home Foundation.
GPL-3.0 - yt-dlp/yt-dlp
A feature-rich command-line audio and video downloader.
Unlicense - Zackriya-Solutions/meetily
A privacy-first AI meeting note taker for Mac and Windows, with on-device transcription.
MIT - blader/humanizer
An agent skill that removes the signs of AI-generated writing from a text, based on Wikipedia's guide to spotting it, and keeps the meaning.
MIT
Saved and worth a look
From my reading list: agents that use a computer or a browser, data from the web, code review, voice, and the self-hosted tools a small business can own.
- vercel-labs/agent-desktop
Computer use for any agent: it reads a desktop app's accessibility tree, so an agent can click and type in it.
Apache-2.0 - browser-use/browser-use
An open-source library for AI agents that drive a real browser, in Python.
MIT - firecrawl/firecrawl
Turns websites into clean markdown or JSON an LLM can read: search, scrape, crawl and map, hosted or self-hosted.
AGPL-3.0 - D4Vinci/Scrapling
A Python web-scraping framework, from a single request to a full crawl.
BSD-3-Clause - alibaba/open-code-review
AI code review that picks and bundles the files deterministically before a model comments on them, line by line.
Apache-2.0 - cloudflare/security-audit-skill
Cloudflare's coding-agent skill for multi-phase security audits, every finding checked by a fresh verifier.
MIT - Mintplex-Labs/anything-llm
Chat with your own documents and run agents, local-first, with any model, including Anthropic's.
MIT - pipecat-ai/pipecat
An open-source Python framework for real-time voice and multimodal agents.
BSD-2-Clause - coollabsio/coolify
A self-hostable alternative to Vercel, Heroku and Netlify: git-push deploys on your own server.
Apache-2.0 - documenso/documenso
The open-source DocuSign alternative, self-hosted or in the cloud.
AGPL-3.0 - gitroomhq/postiz-app
A social media scheduler you can self-host, with an API and connectors agents can use.
AGPL-3.0 - jamiepine/voicebox
An open-source AI voice studio: clone, dictate, create.
MIT - abue-ammar/tinycast
A native, open-source macOS launcher: apps, clipboard, snippets, window actions and a calculator.
Set it up yourself
The setup behind this page, written down. Free, one email.
Want this setup on your team?
I help teams put agents to work in their apps and their delivery: the same tools, with the guardrails around them.