A new way to input text

Throughout modern history, the keyboard has been our primary way of communicating with machines. We type commands, write code, compose messages. But as AI becomes a collaborator rather than just a tool, the amount of context we need to convey has exploded. The keyboard is no longer enough.

Speech-to-text technology has reached a turning point. Speaking is now a viable input method — not a novelty, but a genuine alternative to typing. This is a paradigm shift in how humans interact with computers.

Like learning to type on a QWERTY keyboard, voice input requires practice. But the learning curve is far shorter, and the productivity gains are far greater. We want to provide not just a tool, but a new experience — one that changes how you think about input itself.

Most voice-to-text tools are built in the US, optimized for English. They work well enough for English speakers, but quality drops sharply for other languages. When languages mix — Japanese with English technical terms, Vietnamese with code snippets — existing tools often break down entirely.

The lead developer of koedesk is Japanese, working from Vietnam. This firsthand experience of using voice input across languages and accents shaped every design decision. We noticed what others overlooked: the accuracy gap between English and the rest of the world.

That is why we recommend ElevenLabs Scribe v2 as the default model. It delivers consistently high accuracy across languages — not just English and Japanese, but Vietnamese, Korean, Spanish, and dozens more. koedesk is currently one of the only desktop apps that offers Scribe v2 integration.

koedesk is intentionally simple. Press a key, speak, release. Your words appear where your cursor is. No extra steps, no configuration required to get started.

Under the hood, we are obsessive about performance. The app is built in Rust — the same language used for browser engines and operating systems. Voice activity detection, audio capture, and transcription initiation are all tuned for minimal latency. When you stop speaking, the text appears almost instantly.

Most competing tools use a two-step pipeline: a speech-to-text model for raw transcription, followed by a lightweight LLM to clean up filler words, fix grammar, and reformat the output. We take a different approach. Scribe v2 is accurate enough that LLM post-processing is unnecessary. We trust the model's output directly.

This is a deliberate choice. As speech-to-text models continue to improve, the need for LLM post-processing will disappear. By keeping our architecture simple, koedesk benefits directly from every improvement in the underlying models — without layers of complexity in between.

Guide Inc. is a software development company based in Tokyo, Japan, with a development office in Ho Chi Minh City, Vietnam. Founded in 2021, we specialize in financial technology — building trading platforms and mobile apps for major Japanese financial institutions.

koedesk was born from our own daily work. As developers building complex financial systems, we needed a faster way to communicate with AI coding assistants, write documentation, and compose messages — without taking our hands off the keyboard flow. We built the tool we wished existed, and decided to share it with the world.