You're explaining the settlement logic at 11pm. Two minutes into the voice memo, you've painted the whole architecture: async hooks, the state machine, why you need to batch reconciliation. Then you hit the word limit. Wispr maxes out at 2000 words per day on the free tier. You're at 1847. You have to stop, copy what you've written so far, save it in a draft, and start over in a new session.
This is the problem indie developers and builders have with cloud-based voice-to-text tools.
When the bottleneck shifted from typing speed to explaining intent to LLMs, the tools didn't shift with you. The old voice dictation products, built for meeting notes and email, still price and cap as if transcription is the hard part. It isn't. Transcription is solved. What's hard is writing the long-form prompts, design specs, and PR descriptions that make LLMs useful.
Here's how we got here, and why the free tier matters more than you think.
The workflow changed, the tools didn't
Three years ago, developers mostly typed code. You'd open an editor, type a function signature, and your fingers did the work. Voice tools existed (Dragon NaturallySpeaking, Wispr) but they felt like novelty, something for meeting notes and not real work.
Now LLM-assisted development flipped the equation. You're not typing code anymore. Copilot and Claude Code type code. You're typing intent: a system design doc, a PR description that explains why a function changed, a Cursor chat explaining how two components interact, a Notion spec at midnight when you're thinking clearly.
This kind of writing is different from code. It's verbose, explanatory, context-heavy. It's also faster to speak than to type. A design doc that takes 35 minutes to type takes 12 minutes to speak and clean up.
The problem: cloud-based dictation tools priced themselves for the old use case. Wispr charges $14/month for unlimited. The free tier gets 2000 words per day. Willow is $12/month, free tier capped at 1500 words. Superwhisper is $8.49, also capped. These aren't unreasonable prices for casual meeting notes.
But they're misaligned with how developers actually work now.
The cap is a mental interrupt
Marcus, a backend engineer at a Series B fintech in Stockholm, built his incident postmortems and design docs by voice for a few weeks using a cloud tool before switching back to typing.
"I hit the free-tier word limit three times in two weeks," he said. "Each time, I'd lose the train of thought. I'd have to save what I wrote, start a new document, then stitch everything together the next morning. By that point, the thinking was stale. I was editing prose instead of flowing."
This is the real cost of caps on free dictation. It's not the price. It's the friction of the interrupt.
Developers who work with LLMs spend more time in Slack threads, PR descriptions, design docs, and chat interfaces than they used to. A 2000-word cap feels arbitrary and low. A 1500-word cap is laughable. You're designing a feature. You're describing a bug. You're writing a proposal for how to solve a data consistency problem. The words add up fast.
The tools that cap at 2000 words aren't protecting you or saving money. They're protecting their margin on cloud infrastructure.
Local transcription flips the equation
Recitey runs Whisper locally on your device. Whisper is a 96.3% accurate open-source model trained on 680,000 hours of multilingual audio. The local version means:
- No word limit. None. You can dictate a 10,000-word design doc without seeing a counter.
- No cloud upload. Your code stays on your machine. Your IP, your prompts, your incident context; it never leaves your device.
- No per-device metering. The model runs once. Variable cost is effectively zero after initial installation.
The free tier gives you this. Uncapped. No payment, no monthly subscription, no switching between sessions to reset your daily counter.
The Pro tier adds cloud-side rewriting, a second pass that polishes rough speech output into publication-ready prose. Not everyone needs it. Developers who work in technical channels (Slack, GitHub, Notion, Cursor) often don't. But it exists for people who want the output to be more refined than "this is what I said out loud."
Why this matters for how you work
The shift from typing code to typing intent means voice writing changed from optional to structural for many developers.
When you're in Cursor explaining a complex refactor, you can speak it 3x faster than you can type. Same with shell scripting documentation, API design, or a PR description that explains reasoning (not just the diff).
But only if the tool doesn't interrupt you.
Cloud-capped tools interrupt you. They do it politely, a "word limit reached" message, but they interrupt the flow. That's fine for meeting notes. It's not fine for the kind of writing LLM workflows demand.
Local Whisper doesn't cap. You flow until you're done. Then you move on.
The other thing: developers are allergic to sending proprietary code to third-party cloud services. If your tool requires upload to transcribe, you're asking engineers to weigh convenience against security concerns. Many choose security. They go back to typing.
Recitey's local-first approach solves that tension. Speech-to-text happens on your machine. Your code stays on your machine.
The pricing reflects the technology, not the distribution
Most premium voice tools charge what they charge because distributing cloud services is expensive. Servers, uptime, bandwidth, redundancy. If you cap users at 2000 words per day, you're managing infrastructure costs.
Local Whisper has a different cost structure. You run it once, on your device, with your GPU (or CPU). Marginal cost per word is zero.
That's not altruism. It's a different way to build the product. The free tier reflects that technology choice: no cap, because there's no variable cost to cap.
The Pro tier, with cloud-side rewriting and formatting, is where you'd pay if you want that second pass. But you don't have to. Plenty of developers use local speech-to-text alone and consider the job done.
What this means for you
If you're working with LLMs, you're writing more prompts and specs than you used to. If you hit word limits on free dictation tools, or if you've decided to avoid cloud transcription for privacy reasons, the local Whisper approach changes the equation.
No caps. No uploads. No interrupted flow.
The free tier is the real product. The Pro tier is for the polish.