Your design doc is flowing at 11pm. You're explaining a system behavior to your future self and the engineer who'll review it. Halfway through the third paragraph, the dictation stops. Word cap. You finish typing. The prose doesn't match. You leave it for morning and know you won't rewrite it properly.
This is the developer's voice-writing problem that cloud dictation doesn't solve. The caps aren't there to protect the service. They're there because hosting speech-to-text at scale costs money.
Why developers hit the word cap
Most premium dictation tools cap the free tier around 600 to 1200 words per month. Wispr Flow tops out at 600 on free. Willow caps free at 1000. These aren't privacy-first choices or UX choices. They're pricing mechanics. The service has to recover infrastructure costs, so free tiers get limits.
For developers, this is exactly backwards. A design doc. A detailed PR description. An incident postmortem. These are the moments when voice writing saves the most time. They're 1500 to 4000 words. You're explaining a system behavior, not sending a Slack message.
The local-first difference
Recitey's free tier runs speech-to-text locally on your machine using Whisper. No word cap. No metering. No monthly quota. The reason is straightforward: Whisper runs on your device with zero variable cost after download. The only cost is compute time on your hardware, not hosting time on a server.
This matters because the bottleneck in LLM workflows isn't typing. It's writing the intent clearly enough that the model understands. Marcus, a backend engineer at a Series B fintech in Stockholm, will spend 20 minutes crafting a GitHub PR description that explains settlement behavior clearly for review. Speaking it is faster. But not if you hit the cap halfway through and have to switch back to typing.
Privacy and IP by default
When your design doc contains code patterns, architecture decisions, or business logic, sending that voice data to a cloud service is a trade-off. Some developers accept it. Many don't. Especially in fintech, the default assumption is that code and spec text stay on the device. Marcus refuses cloud transcription for exactly this reason. If the tool won't run the speech-to-text on his machine, he won't use it.
Whisper running locally means your design doc, your bug explanation, your PR comments. They're transcribed on your machine. The only thing that ever leaves your device is the typed-up text you choose to send.
Cursor sketches, LLM intent, voice specs
The workflow shape changed. You used to dictate for quick notes or accessibility. Now you dictate design docs, prompt engineering scratch pads, and intent specs for Cursor or Claude Code. These are long-form, technical, and they need to be accurate enough to feed to a model. Marcus specifically uses Cursor over VS Code because its tab-complete reduces the voice rewrites. But that still doesn't solve the word-cap problem when your design doc is long.
Recitey works in the terminal, in VS Code, in Slack, in browser text boxes, in GitHub comments. Anywhere you can paste text. It's not locked to a single IDE or agent. You draft in your tool, paste it where it belongs, and keep going. The transcription is just the first pass. Recitey includes a local rewrite that cleans up the rough draft in under two seconds, so it sounds like the careful version of yourself, not the rambling version of your voice.
Who this is for
If you're a developer writing design docs by voice because it's faster than typing, and you've hit word caps before, or you've wanted to use voice but avoided cloud services because of IP concerns, this removes both friction points.
If you need monthly metering to control costs, or you need a specific agent integration, or you need phone dictation, other tools fit that better. But if you want uncapped, local-first, privacy-native dictation that works across your entire development toolkit without locking you into a platform, there's no competitor with this combination.
The design doc at 11pm should flow as long as you need it to.