murmr

Speak sloppy, prompt sharp.

Talking to a coding agent well is a writing problem. You know what you want, but typing it out precisely is slow, and saying it out loud produces something too loose to be useful. murmr closes that gap: hold a hotkey, say the messy version, release, and it hands back a structured prompt.

Transcription runs locally through Whisper, so nothing leaves the machine on its way to becoming text.

The design decision that matters

murmr deliberately does not answer the question.

Given “can you help me figure out why the login page is really slow, i think it’s the api calls but not sure, dig into it and fix it”, the obvious move is to start debugging. murmr instead compiles that into a task and its context, and stops:

TASK: Investigate and fix the performance issues on the login page, with focus
on API call optimization.
CONTEXT: The login page is slow; API calls are suspected as the primary
bottleneck but the root cause needs confirmation.

That output goes to whichever agent you were going to use anyway. The point is to make the ask precise, not to take over the work — you stay the driver, and the tool that is good at building prompts stays separate from the tool that is good at writing code.