Marcus is a backend engineer at a Series B fintech in Stockholm, and he spends most of his day in Cursor writing design docs and architectural notes. A few months ago, he started dictating these into voice-to-text while his hands stayed on the keyboard, reviewing code or jumping between pull requests. It worked great until his cloud dictation tool hit its free tier limit mid-design-doc.
He'd dictated maybe 600 words into a sketch for their payment settlement redesign when the tool went silent. No warning. Just a paywall.
Why Voice Became the Bottleneck
The workflow shifted. We don't type code anymore; we type intent for Claude and Cursor to write code. That's a different activity. It demands longer thinking, more complete specs, architectural framing. You're not writing code in prose; you're writing instructions for a model to write code.
That's more words per day than you'd think. Marcus used to type maybe 200 words of internal notes. Now he's dictating 400-word design doc sections, 300-word PR descriptions, 200-word incident postmortems. The bottleneck moved. It's not the typing speed anymore. It's the ability to capture long-form thinking while your hands are on the keyboard.
Voice dictation solved that. Except it doesn't.
The Cloud Trap
Every cloud-based dictation tool caps the free tier. Wispr Flow caps it at around 600 words per month. Superwhisper at 1000. Willow is similar. The paid tiers aren't expensive ($14 per month for Wispr, $8.49 for Superwhisper), but the metering itself creates a permission problem.
When you're counting against a limit, you ration. You ration words. You shorten thoughts. You stop mid-explanation because you've hit 500 of your monthly 600. Then you stop using the tool and go back to typing.
The infrastructure doesn't require per-word metering. Cloud delivery costs money, sure, but the actual transcription is cheap. The caps are about monetization structure, not technical necessity. And they shape behavior exactly as they're designed to: convert free users to paid, or lose them to friction.
When Local Beats Cloud
Whisper (OpenAI's open-source speech-to-text model) runs locally on your machine with the same accuracy as the cloud versions. Whisper-large-v3 achieves 96.3% word accuracy on LibriSpeech, and the model weights are free to download. Running it locally costs nothing per word, because there is no per-word infrastructure. You run it once on your machine.
Recitey implements this. The free tier runs Whisper locally, and there's no word counter because there's no variable cost to meter. Marcus dictated 8,347 words in his first month of using Recitey without ever seeing a paywall or a counter.
The trade-off is real: local Whisper output is rough. It needs polish, punctuation cleanup, maybe a rewrite for tone. Recitey's paid tier does that via cloud rewrite (that's where the variable cost lives, and that's where pricing makes sense). But if you don't care about prose perfection, the free tier is complete. Marcus mostly doesn't use the paid tier. Good enough is actually good enough.
The Rewrite Isn't Free (And Doesn't Need to Be)
This is the model most voice tools miss. They cap free dictation to push you to paid. Recitey uncaps free dictation and monetizes on the rewrite instead.
Free: local Whisper, zero metering, speak as long as you want. Paid: cloud rewrite, polish, tone adjustment, model choice.
For developers who care about code IP leaving their machine, or who optimize for latency, or who just trust their own instincts more than an algorithm's polish, free tier is complete. Recitey works in Slack, email, Cursor, terminal, any Windows app via clipboard. You dictate, you get text, done.
Marcus's Month
Marcus switched from Wispr Flow in April. In thirty days, he dictated 8,347 words: design doc sections, PR descriptions, Slack explanations of bug investigations, incident postmortems, code review threads. He hit none of the friction points he'd hit before.
The integration with Cursor felt native (he specifically chose Cursor over VS Code because tab-complete reduces voice rewrites, and this reinforced that instinct). No hesitation before hitting dictate. No moment where he thought "am I wasting my word budget." Thought flowed, text landed, momentum continued.
He uses the paid rewrite maybe twice a week. Usually he doesn't bother.
The Privilege of No Counting
When a tool doesn't meter you, the permission structure changes. You're not asking "do I have words left." You're just dictating. The friction disappears.
Most tools hide this behind design patterns. Counters, warnings, email reminders that you're approaching the cap. Psychological pressure in the interface. Recitey has no counter because there's nothing to count.
If you've ever hit a word limit mid-thought, you know the shape of the frustration. You know the moment the tool stops being useful and becomes a constraint. Local dictation removes it.