Marcus is three hours deep in a design doc at 11pm. The settlement logic is clear in his head. Voice is faster than typing. Seventeen minutes in, his dictation tool hits its monthly cap and stops listening. The flow breaks. The next morning, the doc is half-finished and the thinking is cold.
He's hit the same ceiling twice before on cloud dictation tools, Wispr caps free users at 2000 words a month, Willow at 1000. Each time, the cost to resume is not the $12 subscription. The cost is context loss. Rewriting the half-finished thought never lands the same way.
The structural problem is not new. The workflow that exposes it is.
How the bottleneck shifted
Writing code used to be the load-bearing task. Developers typed function names, logic, tests. If you spoke faster than you typed, a voice tool was a nice-to-have.
Now the load-bearing task is writing the intent that an LLM needs to build the code. "Here is what the settlement engine should do" becomes a 600-word design doc. "Here is the edge case we hit" becomes a Slack thread that explains the data flow in three paragraphs. The thinking happens out loud first. Typing is the transcription.
Whisper-large-v3 (OpenAI's free model) reaches 96.3% accuracy on standard audio. It runs locally. No API calls. No metering. Zero marginal cost per word. The technical capability is there.
The question is: will your dictation tool let you use it without a word ceiling?
Why local speech-to-text changes the economics
Cloud dictation tools charge because they have cloud costs. Otter.ai transcribes to their servers. That infrastructure costs money. So they meter free users and charge $12/month for unlimited.
Local transcription has no such burden. It runs on your device. The model ships once, and then every word is free.
Recitey transcribes locally. No server, no word cap, no monthly reset. The distinction is not marketing positioning. It is literally what the product does.
This is not an argument that cloud is bad. Cloud transcription can do things local cannot, cross-device sync, search history, team playback. Those are real features. If you need them, cloud is worth $12. But if you don't, the default should be local and free.
The trade-off is real
Local transcription on your device is faster (no network latency) and cheaper (no subscription). It is also less polished. Whisper outputs raw spoken text. "Um" filler words. Incomplete sentences. Capitalization guesses. Fragments.
Recitey's paid tier ($14/month) handles the rewrite: polish, punctuation, structure. It takes the raw dictation and smooths it into a clean sentence that you can paste into Slack or a PR without editing.
The free tier gives you the raw Whisper output. You can use it as-is or clean it yourself. The choice is yours.
The point is: you get the dictation without a ceiling. You do not get the polish unless you pay.
For the workflow Marcus actually runs
Marcus works in Cursor, not VS Code. He switched to Cursor specifically because its tab-complete feature reduces voice rewrites. He refuses cloud transcription because code IP concerns him, design docs, PR descriptions, incident postmortems all contain company data that should not leave his device.
For Marcus, local dictation is not a feature. It is a requirement.
Local, uncapped free tier meets that requirement. He can voice a 1500-word design doc at 11pm without hitting a word counter. The thinking stays intact. The next morning, the doc is complete. If he wants the prose polished, he subscribes for the month. If he does not, the raw dictation is useful as-is.
This is not a use case every developer has. Many developers prefer typing. Many use cloud tools and do not worry about code IP in transit. For those developers, another tool may fit better.
But for the developers whose workflow depends on long-form voice dictation without a meter, the word cap is not a limitation to tolerate. It is the thing that breaks the work.