The workflow shifted. You're not writing code anymore. You're writing instructions for code.
PR descriptions that explain intent before the diff. Design docs at 11pm that need to be in your head before the model can build it right. Slack investigations that walk through a production incident step by step. All of that takes time to type. Voice is faster. But voice dictation tools have word caps, and hitting that ceiling mid-thought breaks your flow.
The Cap Hits Harder Than It Looks
Wispr Flow gives you 2,000 words a month on the free tier. Superwhisper caps you at 600 minutes a month. Willow limits you to 1,000 words. That sounds reasonable until you're mid-thought in a design doc and you hit the ceiling mid-sentence.
Marcus, a backend engineer at a fintech in Stockholm, does this regularly. He'll start dictating a design doc at 11pm, structure three sections cleanly, and then the counter clicks over. He's forced to stop, pay to continue, or paste the draft into a text editor and finish typing. The flow breaks. The thinking gets fragmented. He spends the next morning stitching it back together.
"I refuse to use cloud transcription anyway," he says. "I'm dictating payment settlement logic. That code stays on my device."
That's the unspoken cost of cloud-based dictation: someone else's server touches your words.
Recitey Works Locally, Costs You Nothing per Word
Recitey uses Whisper, the open speech-to-text model from OpenAI. It runs on your device, not in the cloud. Your design docs, your Slack messages, your incident postmortems never leave your laptop.
No word counter. No monthly ceiling. No variable cost. You can dictate 10,000 words if you want.
The model accuracy is 96.3% on LibriSpeech, the standard benchmark. That's close enough for first drafts. You clean up the 3.7% in the rewrite.
The rewrite is where Recitey's cloud layer comes in, if you want it. The free tier gives you the dictation. The Pro tier gives you the polish: a sentence rewrite that takes your rough voice output and makes it sound like you had typed it. But the microphone part runs offline.
Why This Matters for Developers
You work in Cursor, Claude Code, GitHub, Slack. Those tools all have different input models. A voice tool that only works in one IDE or one chat interface is a toy.
Recitey works through your system clipboard. Speak. Paste into Cursor, Slack, a Notion page, a Linear comment, a GitHub PR description. The tool stays invisible. It doesn't matter which application you're in.
Compare that to Wispr Flow ($14/month, 2,000-word ceiling) or Superwhisper ($8.49 one-time, 600-minute monthly limit). Both charge you for words or time. Both run in the cloud. Both assume you're dictating customer support responses or casual notes, not payment settlement specs.
The new development workflow is different. The tool should be different.
The Technical Detail You Actually Care About
Whisper-large-v3, the model Recitey runs locally, weighs 1.5GB on disk. It runs on CPU or GPU depending on your hardware. No API call. No network round trip. No logging. Your words never route through a third-party endpoint.
That matters when you're dictating infrastructure code at midnight and you want to know: where did that recording go?
Answer: nowhere. It stays on your machine.
What Stays True
Cloud-based tools have their place. If you need live transcription for a support call, or real-time collaboration where someone thousands of miles away hears your words as you speak, cloud makes sense.
For the work Marcus does, deep-thinking design docs, incident postmortems, detailed PR descriptions, those can wait 30 seconds for local processing. The latency cost is zero. The privacy cost is gone. The word-count anxiety disappears.
The free tier has no ceiling. That's not a marketing claim. That's the technical structure: the model runs on your device, transcription costs nothing, metering is not how the business works.