Pharmacist · Angler · Builder of local-first desktop AI
From prescriptions to pull requests
Local-first by conviction. Private, fast, honest software that runs entirely on your machine — no account, no telemetry, no server, nothing sent anywhere. Every app below can be checked by reading its source.
| 3 | 5 | 160M | 0 |
|---|---|---|---|
| apps shipped | repos open source | tokens, cache excluded | servers required |
I am a power user of AI coding agents, and I treat it as a discipline rather than a shortcut.
160 million measured tokens of real work sit behind the apps below — cache reads excluded, so it is the work itself and not the same context counted twice. That number is measured rather than guessed, because I built and released the meter that counts it. The agents write; I direct, review and verify. Every product decision, every architecture call and every security model in these apps is mine.
The method is written down and open-sourced as Hash AI Coding Persona — persistent memory, git work-trees, signed-commit discipline, and verify-before-you-trust. It is how you move at agent speed without the mess.
I do not reach for orchestration frameworks. Everything below — the swarms, the planners, the tool registries, the retrieval — I designed and had built from first principles.
|
The local-first AI workspace for macOS. Ten workspaces — Coder, Agent Swarm, Finance, 3D Forge, Sandbox, Virtual OS and more. 12 providers, or fully offline with Ollama. Nine specialist agents, real Python in a sandbox. No backend, no account. 8.9 MB.
|
The Dynamic Island every Mac deserves. What is playing, internet speed, battery, how hot the chip runs, what you spent on AI today — a glance away at the notch. macOS itself is asked what is playing, so an app nobody wrote support for still works on day one.
|
The honest usage meter for AI coding tools. Counts real measured tokens, not estimates — across six tools. No server, and it never reads your chats. It is the meter behind the 160M above.
|
|
A local-first medical-research workbench — a 3D anatomical brain that searches, cites and peer-reviews the literature. A query planner decomposes the question, specialist agents collaborate over a shared blackboard, and every claim is graded against the evidence it cites before you see it.
|
A living fishing atlas on a 3D globe, with an AI fish specialist that flies the earth to show migrations and catch spots — an agent driving a real interface through tool calls, not returning text. (The angler in me, in code.)
|
The way I work, and the domain knowledge behind it — written down so anyone can take it.
|
The written-down system for shipping serious software with AI coding agents — persistent memory, git work-trees, signed commits, verify-before-you-trust. Copy-paste templates. Works with any agent, any model, including local Ollama.
|
Drop-in agent skills that make an AI appraise medical literature like a reviewer — grade every citation against its source, run reproducible PRISMA searches, pool studies safely, and defend retrieval against prompt injection.
|
Nothing about you ever leaves your machine. · Every commit signed. · Every claim checkable.