Two of you. One does the boring part.

Twynn is in active development, not yet shipped. It is designed as your digital twin, a small desktop companion meant to hold the desk while you are away and, over time, learn the shape of how you work, all of it processed on the device rather than in the cloud.

The problem it targets

The desk rarely empties: routine clicks, scrolls, and app-hops fill the day without needing full attention, and looking active can matter as much as the work itself. Robotic automation gets caught quickly, macros and bots move in dead-straight lines at inhuman speed and are easy to flag. Context about where the hours actually went tends to live only in your head, workflows repeated fifty times a week become invisible because they're automatic, and most monitoring software ships your activity to someone else's server.

Planned capabilities

• Human-like input. Bezier mouse paths, realistic speed, and micro-tremors, aiming to move like a person rather than a script.
• Workflow simulation. Chaining app switches, scrolling, keypresses, and tab switches into sequences meant to read as real work.
• Three activity levels. Low, medium, high, to match the rhythm a given role calls for.
• Auto Mode. Designed to notice when you step away, around 30 seconds idle, take over, and stop the instant you touch anything.
• AI Mode. Intended to learn per-app patterns and replay them contextually.
• Stealth. Designed to stay invisible to monitoring tools and task managers, with the option to present as an ordinary application.
• Insights. App usage, sites visited, and a productivity breakdown surfaced from what the twin did while you were out.

Engineering challenges

• Making a cursor human. Real hands move in curves with imperfect timing; synthesizing bezier paths and micro-tremors that read as human, without looking scripted, is the core design problem.
• Staying invisible. Being undetectable to monitoring tools and task managers requires careful process design and the option to present as an ordinary app.
• Intelligence in a small footprint. Running observation, clustering, and a memory model on-device within a couple of megabytes forces ruthless choices about what stays resident.
• A clean hand-back. Auto Mode needs to detect real presence and yield the moment you touch the machine, with no lag and no fight over the cursor.
• One behaviour, three operating systems. Native input control differs across macOS, Windows, and Linux, and a single codebase has to feel identical on all three.
• Private by construction. The memory is valuable precisely because it's sensitive, so the design keeps it off the network entirely, as a constraint rather than a setting.

My role

As founder and product architect, I set the product vision and its two founding convictions: move like a human, and stay on the device. I designed the architecture, a single desktop process with an automation engine and an on-device intelligence layer, and I'm building the human-motion engine (bezier paths, speed curves, micro-tremors) and the observe, learn, remember, understand pipeline that is meant to run entirely offline. I also own the brand, the marketing site, and the product's voice, and I'm driving the roadmap from concept through private development toward a first release.

Technology

The desktop target spans macOS, Windows, and Linux from a single binary. Automation is built on native input APIs, motion synthesis, and idle detection. The intelligence layer is on-device processing, event clustering, and local models, with a local, encrypted-at-rest memory store and no cloud dependency. The design budget is roughly 2MB of RAM and a sub-5MB download, offline-first throughout.

Where it stands

Twynn has moved from concept through research into what reads as human motion, then into designing an on-device-first architecture, and is now in private development, building the engine and intelligence to fit its footprint target across three operating systems ahead of a first release.