It's 11pm on a Tuesday. Marcus is in Cursor, working through the design of a new settlement flow. He's got the logic clear in his head, the async retry pattern, the idempotency window, the state machine for handling chargebacks. He starts dictating into his microphone.
Two minutes in, at about 800 words, the dictation stops. Word cap hit.
He switches to typing. The prose fractures. What was flowing explanation becomes fragmented bullets. He loses the narrative thread. By morning, he'll have to rewrite it anyway.
This moment is the dirty secret behind most voice dictation pricing.
The cap isn't a technical limit
It feels like one. You hit a hard ceiling. The system stops listening. You'd think the software just can't handle it.
But Whisper, the underlying model that powers Recitey, runs locally on your device. No API call. No server involved. There's no technical reason to count the words.
The cap exists because other tools have chosen to meter it.
Wispr Flow caps the free tier at 2,000 words per month. Superwhisper at 2,000 words per day. The technical cost per word is zero. The business model cost is everything.
Most SaaS pricing reflects distribution and operations overhead, not the actual technology. A local Whisper model has neither. It runs on your machine. It costs nothing to operate at scale.
What changes when there's no meter
Without a cap, Marcus finishes the design doc in one voice session. The thinking stays coherent. The prose has continuity even on the first pass.
He still runs it through the Claude Code rewrite engine in Recitey Pro, the cloud-side that polishes rough transcription into clean prose. But the raw capture is complete. The thought is whole.
That's the real workflow. Dictate the full intent, then polish it. Not: dictate until stopped, then fragment it across sessions.
The cap doesn't protect you from bad transcription. Whisper-large handles technical terms and domain specificity well enough for a backend engineer explaining settlement logic. The cap just breaks your thinking into artificial pieces.
The local part isn't paranoia
Marcus refuses cloud transcription tools. Fintech, IP concerns, code samples in design docs. He uses Cursor specifically because its tab-complete reduces the need for cloud rewrites.
Recitey's free tier works locally. Whisper runs on device. Nothing leaves your machine except what you choose to send for the Pro rewrite service.
That's not a feature. That's the baseline.
The pricing shape that makes sense
The free tier is unlimited local dictation. You get the thinking tool, unmetered.
Pro is optional. It's for the cloud-side rewrite engine if you want AI to clean up the transcription. Not subscription-level. Not a tax on raw voice input.
This is backwards from every other voice tool. They meter the capture. You pay to think.
Recitey meters the polish. You pay if you want it cleaner. Thinking is free.
For developers building on intent
The workflow changed. Code used to be the bottleneck. Now it's specification.
Marcus writes more words than he types. Not because voice is faster for everything. Because explaining intent to an LLM requires clarity that keyboard typing struggles with. Voice captures the thinking. The model clarifies the intent.
A word cap mid-intent is a cap on your model's ability to receive it.
The local Whisper model and the no-cap policy aren't marketing differentiators. They're structural acknowledgments that the bottleneck has moved. Developers need to think out loud. That shouldn't have a metered ceiling.