Christopher Staats · Independent Developer
Four apps live on the App Store. A native desktop app for Mac and Windows that runs its AI locally — and where general-purpose models weren't good enough, I trained my own. This page is hand-written HTML, CSS, and JavaScript — no frameworks, no build step. That's the point.
Four apps live on the App Store, a desktop app on three platforms, and trained models running on-device.
Flagship
One codebase. Three native platforms. AI that never leaves the machine.
FileComb started as a web app for working with documents, then became a native desktop app built with Tauri — shipping on Intel Macs, Apple Silicon, and Windows from a single codebase. The AI, including vector search across your files, runs locally on-device, so file contents never leave the machine.
An iOS IDE with an agentic AI coder — powered by local models.
pocket/grammer is a native iOS IDE where an AI agent plans, writes, and revises code step by step, with per-step review and undo. Models run locally on-device via llama.cpp with Metal GPU acceleration — optional bring-your-own-key cloud providers, and the phone can even pair with your computer to serve a full desktop coding UI. Designed, built, and shipped end to end, from empty project to a product anyone can download.
An iOS news app with an on-device trained AI — and the marketing site that ships with it.
A native iOS news app, live on the App Store, built alongside its own marketing website. Two surfaces, one product: the app people install and the site that introduces it. Its newscasts are written on-device by a small LLM I trained specifically for the job — the same pipeline behind Podkipedia's generator.
AI timestamps for gameplay footage — on iPhone, iPad, and Mac.
Killstamp finds the moments that matter in gameplay recordings: tap a UI element on three frames, and it scans the footage with on-device template matching, logging frame-perfect timestamps with automatic clip creation and CSV/JSON export. No cloud, no upload — and a one-time purchase, no subscription.
Type any topic. Get a two-host AI podcast — generated on your iPhone.
Podkipedia researches any Wikipedia topic and produces an hour-long, two-host episode — curious Ava asking, expert Jordan explaining — written in stages and narrated with on-device text-to-speech. Playback streams while later segments are still being written, and everything stays on the phone: no server, no account, no data collected. Free, no subscription.
Book clients. Send invoices. Own your data.
BookKit is a native appointment-booking and client-management app for iPhone, iPad, and Mac: an availability engine that prevents double-bookings, a shareable booking link that works without an app or account, self-numbering invoices with PDF export, and OCR camera scanning for paper invoices. Runs entirely on-device — no account, no cloud, no subscription. Launching on the App Store soon.
Small models trained for one job, running inside the apps above.
A distillation pipeline that trained Qwen3 0.6B/1.7B to replace general-purpose LLMs as the on-device generators in two shipping iOS apps: teacher-student data generation, LoRA training with MLX on Apple Silicon, GGUF quantization, and results measured against each app's own quality gates — the 1.7B newscast model went from 20% to 100% gate-pass.
The tools I ship with, grouped by where they run.
Swift · UIKit + Objective-C interop · SwiftUI · App Store distribution (4 live apps)
Tauri · macOS code signing + notarization (Developer ID) · Windows installers
TypeScript · React · Node.js · PostgreSQL · Drizzle
LLM training (LoRA, MLX on Apple Silicon) · GGUF quantization · On-device/local inference · RAG + vector search (Qdrant) · LLM API integration · Prompt architecture with guardrails
I'm an independent developer who ships complete products end to end — design, build, test, ship. For the last two years I've worked with an AI-native workflow, using AI-assisted development on top of strong fundamentals to move at a pace that used to need a team. It's how one person ships a desktop app to three platforms or takes an iOS app from idea to the App Store.
This site is part of the pitch: hand-written HTML, CSS, and JavaScript, with no frameworks and no build step. Simple things, finished.