You're writing a design doc for the payment settlement feature at 11pm. The system design is clear in your head: auth flow, webhook retry logic, ledger state machine, idempotency keys. You start dictating. Three minutes in, you've got 800 words. Five minutes in, you hit the word limit on your voice tool. The transcription stops. By tomorrow morning, your train of thought is fragmented across three separate drafts.
This is the tax most developers pay with cloud-based voice dictation. Wispr Flow caps free users at 600 words per month. Willow caps at 1,000. Superwhisper at 2,000 per month. These limits exist because the cloud model charges per minute of inference. The economics push toward metering.
Recitey doesn't have that constraint. The speech-to-text engine (Whisper, running locally on your machine) has no variable cost. No monthly cap. No word meter. You can dictate a 4,000-word design doc from start to finish without interruption.
This changes the shape of what voice writing actually enables.
The Prompt Writing Bottleneck
When you shifted from typing code to typing intent, explaining what you want the model to build, not building it yourself, the bottleneck moved. It's no longer typing speed. It's specification clarity. You're writing more prose than you ever did before: design docs, PR descriptions, Slack explanations of architectural decisions, incident postmortems, detailed issue reproductions.
Typing gets exhausting at that scale. Your hands hurt. Your brain is already spent thinking through the architecture. But you still need to write it down, clearly, or the next person (or the model, or you six weeks later) won't understand the decision.
Voice should help here. But only if the tool doesn't interrupt mid-thought.
Why Local Matters for Code
There's also a secondary reason developers trust local Whisper: code doesn't leave your device. When you're dictating design docs or PR descriptions that reference actual code, IP concerns aren't abstract. A fintech backend engineer shouldn't be uploading audio of payment logic to a third-party cloud service, even for transcription.
Recitey runs the transcription locally. Your audio never leaves your machine. Neither does your code context.
The Uncapped Part
The uncapped part is structural, not a concession. Because the variable cost is zero, there's no business reason to impose artificial limits. You can dictate as much as you want, whenever you want.
The paid tier (Pro) isn't for more dictation. It's for the cloud-based rewrite engine, the part that polishes your rough voice draft into a clean sentence with proper structure. That lives in the cloud because it's expensive to run and you might not need it every time. But the raw dictation? That's local and unlimited.
Marcus, the backend engineer in Stockholm, uses Cursor instead of VS Code specifically because Cursor's tab-complete reduces the number of times he has to rewrite voice output by hand. The moment he realized he could dictate design docs without hitting a monthly cap, he stopped seeing voice writing as a gimmick and started using it for serious spec work.
The Workflow Difference
This shifts the voice writing experience from "use voice for short bursts, finish in text" to "use voice for the full thought, edit in text if you want."
Design doc at 11pm? Dictate the whole thing. Three thousand words. No cap, no interruption, no IP concerns.
PR description that explains why you chose event sourcing over CQRS for the settlement system? Dictate it in one go.
Slack thread explaining a subtle bug in the webhook retry logic? Same thing.
The tool gets out of your way instead of imposing artificial constraints that force you to context-switch back to the keyboard. The moment you stop thinking about limits, voice writing becomes part of the design process instead of a side tool.