Harness

For Mac and iPhone, free and open source

Three replies in a row without touching the keyboard: Fn, say it, and it's pasted where the cursor is. Harness keeps every take, so the one that came out blunt can be reworked before it goes.

Harness is a Mac app I made for myself. It started as my attempt to improve on Wispr Flow and other voice dictation tools, and it's grown into a full agent harness: chat, notes, tasks, memory, and tools, with an iPhone app for capturing away from the desk.

It's my test bed for new ideas and interaction patterns. Just as important, all the data is mine. It lives on my Mac, and I can see exactly how it's used and when it isn't.

If you want to try it, download the latest build (Apple silicon, macOS 12 or later) or read the source.

Dictation

Press Fn from any app and say what you're thinking. Apple Speech transcribes it on the Mac, it lands in a conversation, and the reply streams back.

It's built to run for a long time, so I can talk for minutes on end without worrying about losing anything. Nothing in the pipeline is a black box, either. The audio file for every take is saved, even when transcription fails, and I can rerun the transcription on it. The prompt that cleans up the transcript afterward is editable, as is a list of names and terms it should spell my way.

A new chat with a recording pill at the top of the window showing a live level meter, the elapsed time, and buttons to cancel or finish.
Mid-take. The pill shows the level and the time, and stays out of the way.

On the phone

The iPhone app does one thing first: record. Open it and it's already listening. The home screen is a single circle, and a lock screen widget opens straight into a take.

A finished take is saved as a capture and synced to the Mac, along with chats and tasks, through a Cloudflare R2 bucket that I own. The phone app isn't on the App Store. It builds from the same repo with Xcode.

The iPhone home screen of Harness: a single white circle on black, with a small menu button at the bottom. Harness on iPhone while recording: a timer at the top, a field of dots that moves with the voice, and a stop button.
The whole home screen, and what it turns into when I start talking.

Grill mode

Ask to be grilled and it interviews you instead of answering, one question at a time, with the choices in the composer. I reach for it when the hard part is knowing what I think.

A Grill conversation: a short reply ending in a question, with three lettered answers in the composer and room to answer in your own words.
Each answer leads to the next question.

Notes

Chat stays short. Longer writing goes into a Markdown note that the model can draft into or suggest edits to, and that I keep editing by hand.

A Markdown note titled Week of September 14, with decisions and open questions, beside the library.
A weekly note with decisions and open questions.

Tasks

Ask in chat or dictate, and tasks land in a plain list. There's nothing to set up and no project hierarchy.

The task list: one task under Now, three more active, and two completed.
The task list, with a couple of things already done.