You're 45 minutes into dictating a design doc. The architecture is still flowing. Then the app tells you you've hit the free tier word cap. You lose momentum. The next morning, you re-read what you dictated and the prose feels fragmented because the thought got cut off. You spend another hour rewriting it into something coherent.
This happens on every tool with a free tier that caps you.
Why word limits exist
Cloud-based transcription costs money per word processed. Superwhisper, Wispr, Willow, and most others meter their free tier to control infrastructure costs. They're not being evil. It's just economics. If transcription didn't have a variable cost, they would not need to cap it.
The cap is a proxy for "we pay for this, so you need to either accept limits or pay us."
Why Recitey's free tier has no cap
Recitey runs Whisper transcription on your machine, not in the cloud. Whisper-large-v3 achieves 96.3% word accuracy on LibriSpeech and runs on modern CPUs at under 500ms per minute of audio. Because the computation happens locally, there's no per-word infrastructure cost to meter. No cost per word means no reason to cap the free tier.
Free dictation, unlimited length.
The trade-off: where the intelligence lives
Running Whisper locally solves transcription. It does not solve rewriting. The raw output is often garbled, grammatically shaky, or contextually wrong. The expensive part is rewriting that rough draft into a clean sentence. That rewrite requires a larger model and typically lives in the cloud. That's the Pro feature.
The architecture is inverted from other tools. You pay for the part that requires real computation. You get the part that is cheap for free.
The unspoken benefit: code IP stays on your machine
If you're dictating code, architecture decisions, or walking through a bug in voice, nothing leaves your device during transcription. For engineers at companies with IP concerns, this is not a small thing. You can draft highly technical docs in voice without uploading them to Wispr's servers or Superwhisper's infrastructure.
Marcus, a backend engineer at a Series B fintech in Stockholm, refused to use cloud dictation tools for exactly this reason. His design docs contain payment settlement logic and infrastructure details he is not comfortable uploading to a third-party cloud. Local transcription, to him, is not a feature. It's a requirement.
How this changes the 11pm workflow
Late-night design docs are common in engineering. You have an idea, you're at your desk, and voice is faster than typing if you're explaining something complex. The problem emerges when you hit the cap. You lose the thread. The next morning, the writing needs repair.
With no word limit, you can dictate the entire thought in one session. You get a rough draft that is long, but coherent. Rewrite happens after. The thinking survives intact.
No word counter means no artificial break in the thinking. That changes the whole experience of voice-first drafting.