You've explained it perfectly on the call. Your payment settlement logic is bulletproof. But now you need to write a spec that the LLM can actually build from, and typing at 11pm after a long day feels slower than thinking.
Voice is faster. But every cloud transcription tool stops you partway through.
The Prompt Is Now the Code
Three years ago, the bottleneck was the keyboard. You wrote code. Something's shifted. Now you're writing prompts, design specs, PR descriptions, incident postmortems. The work moved from "write the code" to "describe what the code should do so the LLM builds it." Marcus, a backend engineer at a Stockholm fintech, runs into this every time he documents a payment settlement bug. He speaks his analysis into Notion and the longer he can keep talking, the more thorough the spec. The tool's job is to get out of the way.
Where Cloud Dictation Falls Apart
Wispr charges $14 a month and caps the free tier at 600 words. Willow's $12 with the same limit. Superwhisper is $8.49, capped. Every one of them stops you mid-thought because the cloud API costs them variable money per transcribed word. They're metering you to protect their margins. When Marcus hits the word cap mid-design-doc, he has to stop, paste what he's got into an editor, start a new file, and the thread of thinking breaks. The spec ends up fragmented. He cleans it up the next morning.
Local Whisper Changes the Equation
Recitey runs Whisper locally on your device. Whisper large-v3 hits 96.3% word accuracy on LibriSpeech. No cloud. No variable cost per word. No metering. You get a free tier that's literally uncapped. Dictate a 2000-word design spec. Dictate a 3000-word incident postmortem. There's no counter. There's no "upgrade to remove the limit." It just works.
The local approach costs Recitey zero dollars per word transcribed. The bottleneck shifted from "can I afford to keep dictating" to "can my device process it fast enough." Whisper v3 handles roughly 12 words per second on modern hardware. That's a couple seconds of latency, which feels normal when you're thinking at speaking pace.
The Trade-off You'll Actually Notice
Local processing is slower than cloud. You speak and there's a beat before the text appears. Cloud would give you that text in 300 milliseconds. But here's what kills design docs more than latency: the word cap. Hitting the limit mid-explanation because cloud tools need to meter you. You can feel a 2-second transcription lag; you can feel a fragmented spec that you'll rewrite tomorrow night.
What This Changes for Your Workflow
Marcus refuses cloud transcription because payment settlement code is IP he's not comfortable uploading, even encrypted. He's also skeptical of tools that lock him into a single IDE. Recitey works in Slack, email, your browser, GitHub, Notion, Cursor (where his tab-complete reduces rewrites), everywhere. His design docs stay on device. His IP stays on device. He's dictating the full thought instead of stopping at a word cap.
Who This Solves For
This is for developers writing specs, docs, PR reviews, incident postmortems. Anyone skeptical of cloud transcription for IP or privacy. Anyone building in Cursor and Copilot and Claude Code, tired of tools that still think you type code for a living. Anyone who's moved from "write the code" to "describe what it should do" and needed the tools to catch up.
The bottleneck isn't your keyboard speed anymore. It's whether your tool understands the new shape of the work.