Rogers

Problem

I wanted a faster way to tell my computer what needed to happen.

I noticed that typing slowed down my productive potential. When typing, the keyboard forced me to compress my intent down into something shorter, where my output was subconsciously (or consciously) mediated by time and effort.

But I could speak far faster than I could type. Also, agents didn’t need my input to be perfectly formatted to understand what I wanted, since they could infer meaning very well from a loosely structured, spoken-out thought just as well as from a tightly edited sentence.

So I found that the fastest way to get good results from an agent was to just talk to it.

Voice mode in existing assistants didn’t solve this, because it was built around an ongoing, live back-and-forth. Since a spoken reply from the agent took longer to listen to than to read, those assistants kept their responses short, which capped how much the agent could actually do or explain per turn. What I wanted was the inverse: the raw speed of speech on my end, paired with a comprehensive written response on the agent’s end.

With reliable cross-app dictation, the iterative loop between me and my agents got faster on both sides at once, and remained high-fidelity.

Solution

Rogers is a floating action button (FAB) that sits quietly on top of whatever I’m already doing on the desktop.

It’s a small floating button that’s always in reach, and is triggered by a single keyboard hotkey.

Dictation produces text wherever I’m working, and a library of my prompt chains and commands sits alongside it, ready to be dropped in without being typed and retyped.

There’s a matching keyboard for the phone too, so the same benefits apply when AFK.

Features

Tapping a hotkey, or clicking the FAB, starts recording, and tapping it again stops and hands the audio off to be transcribed. The output is immediately pasted into the currently selected text field.

Right-clicking the FAB opens a prompt library, a searchable list of prompts and commands split into category tabs.

Workflows are also represented in a visual map of my development workflow. I can find and reuse the exact wording I want for a recurring task instead of retyping it. I always know where I am in the sequence from an initial idea to a finished, documented feature. I can jump in any time to intervene, which is helpful even if using an agent’s skills.

On iOS, the experience is identical to WisperFlow (an app/keyboard pairing that allows me to dictate by clicking a button, triggering an app to record and transcribe). Due to iOS limitations, this is the only workable (and still not ideal) way to work dictation on the device.

Method

Rogers is implemented in Python using the Qt UI toolkit that runs continuously in the background rather than as a program that I’d open and close.

It registers a system-wide shortcut so it can respond from any window, and does its listening and transcribing on separate background threads so it never freezes or blocks whatever else is running.

When the shortcut fires, it captures the microphone, sends the resulting audio out to a speech-to-text model for transcription, and simulates a paste into whatever text field has an active cursor.

The main considerations were reliability at the operating-system boundary so I’d keep text insertion working consistently across very different apps.

The whole thing is packaged with PyInstaller into a single self-contained executable, so it can be run and updated without a separate install step.

The prompt library is deliberately kept as plain files on disk rather than a database, so it stays easy to inspect, edit, and carry between machines, and the visual map is a thin layer that reads that same file structure and lays it out as a connected diagram, with no separate data store of its own.

The iOS version is a separate native Swift app with its own custom keyboard extension, sharing a private app group with a small companion app so the same API key and transcription pipeline carry over without any manual setup — just like WisperFlow.

Reverence

The name is a nod to Carl Rogers, the psychologist behind person-centered therapy. Rogers is remembered for his conviction that empathic and non-judgmental listening is itself the “intervention” that helps a person find their own way toward their goals. Rogers believed people carry their own drive toward growth, and that the listener’s job isn’t to direct them but to listen and accept them closely enough to let that drive do the work.