Independent projects

Software I’ve made with AI, from everyday utilities to local model tools.

Explore the projects

01 / The collection

51 projects and studies.

51 in this view

Photograph junk mail to identify the sender and prepare an opt-out request.

Visit site

Check your domain’s email settings and see which authentication records need attention.

Visit site

Turn a web page into readable Markdown for an AI assistant or your own notes.

Visit site

Keep a Mac awake while AI coding jobs run, then let it sleep when the work finishes.

View source

Reduce Mac cursor lag caused by gaming mice that send movement updates too frequently.

View source

Control extra brightness on supported XDR Mac displays with the usual brightness keys.

View source

Extract text from PDFs so you can search it, edit it, or give it to an AI assistant.

View source

Get several AI models to critique and refine an answer before you use it.

Visit site

Set up a coding assistant that runs on your own Mac or RTX 5090 PC.

View release

Meet other readers through bookstore events and shared books, with conversation starting only after both people agree.

View release

Speak instead of typing in Android apps, with watch and Mac companions for capturing clips.

Help a child practice reading with one short, scripted phonics lesson a day.

Visit site

Listen to podcasts from your own storage and adjust playback speed precisely.

Print downloaded or photographed shipping labels at the right size.

Check your remaining Codex and Claude usage from a watch.

Get a warning when heavy cellular use and poor signal coincide with a hot or power-hungry Android phone.

Search word origins, languages, and dates in the dictionary already installed on your Mac.

Talk to a homemade smart speaker for AI conversation and music around the house.

See wearable health measurements together on an Android phone.

Schedule a Southwest flight check-in from a browser.

Visit site

Give an AI assistant concrete rules and examples for writing and designing with fewer generic defaults.

Visit site

Explore counterarguments to an idea and trace them back to sources.

Visit site

Develop advertising ideas from an offer and the objections a customer might have.

Visit site

Prepare heat-pump job documents for a rebate application.

Visit site

Set a spending limit for an AI task, follow its progress, and review the saved result.

Visit site

Test interest in a product idea with a landing page, signup form, and conversion tracking.

Visit site

Review a proposed product action and keep a record of the decision and its reasons.

Visit site

Review an app’s files for security issues and collect the findings in a report.

Visit site

Check a website for potential security issues and review suggested fixes.

Visit site

Arrange a security review of a website you own or are authorized to test.

Visit site

Help bank reviewers connect credit and compliance decisions to the documents behind them.

List an item for resale from a photo, with support for buyers using shopping agents.

A book manuscript about testing whether a product idea is worth pursuing when AI makes building faster.

Create a children’s story around a chosen theme and a parent-managed profile.

Explore a simple way to browse deals on highly rated products.

Keep local coding-agent conversations, saved sessions, and access rules in one place.

View source

Stop an OpenClaw agent and keep it stopped until you choose to resume.

Watch a local AI assistant work through a browser task one visible step at a time.

Connect DeepSeek Harness to local Qwen models and check whether its tool calls work.

View source

Chat with models running in LM Studio from a browser that also fits a phone.

View source

Publish projects from owned hardware and return to an earlier release when a change goes wrong.

Visit site

Measure how PingPong handles interactive puzzle tasks and review the moves behind each score.

Play a three-lane rhythm game with a 3D stage that moves with the music.

Watch software agents develop ideas, evaluate their work, and branch into new generations.

Explore an early chat interface with demonstrative replies.

Visit site

Early experiments in putting AI chat into a hosted web interface.

An experiment in tracking GPU deals and stock in one place.

Visit site

A concept for putting distance between yourself and distracting phone apps using a physical pass card.

Visit site

A concept for detecting nearby drones by their sound.

Visit site

A concept for helping two businesses work through a dispute with guided offers and an agreement record.

Visit site

A prototype for mapping and organizing a venture plan.

Visit site
Public sites, open tools, prototypes, and earlier studies.The project index

02 / Cloud + local

The tools behind the projects.

70B+

tokens across cloud and local AI

3 × RTX 5090Local AI fleet
450 tokens / secCombined local throughput
~9.7B local tokensEstimated since January 2026
Behind the numbers
Codex61.4BLifetime dashboard counter
Claude Code~2.5BRetained usage records
Local fleet~9.7BThroughput estimate

Snapshot: September 7, 2026. Codex uses the lifetime counter reported from the app dashboard. Claude covers retained records from January 28 through September 6, with duplicated history removed.

The local estimate assumes the three 5090s together sustain 450 tokens per second continuously from January 1 to this snapshot: elapsed seconds × 450. That is about 9.7 billion generated tokens.

Cloud counters include input, output, and reused context. Local throughput estimates generated output. The headline rounds the combined activity down; it is not a count of newly generated text alone.

View the breakdown ↗