Murmr started from a simple need: most dictation tools depended on cloud pipelines and recurring subscriptions. I wanted capture and refinement to run on-device, with output that could go straight into Obsidian, GitHub, or a prompt window without extra steps.
It’s not just transcription. Raw speech is messy; the value is in cleaning it and shaping it into whatever format your workflow expects. So Murmr adds templates and structure on top of local capture and refinement, while staying minimal and privacy-first.
Apple’s on-device models handle both speech-to-text and refinement. That keeps the pipeline local, avoids sending voice to third parties, and fits how I prefer to work: less dependency on external services, more control over where the text ends up.