← BlogFor developers

No Meter, No Interruptions

Marcus is a backend engineer at a Series B fintech in Stockholm. It's 11 p.m., the office is quiet, his team is offline. He settles in to dictate a design doc on the payment reconciliation system he just debugged. The details are still fresh. His thinking is crystallized. Dictation is faster than typing eight paragraphs of technical prose.

He starts speaking. Two minutes in, he's building momentum. He's explaining the idempotency key strategy, the timeout handling around failed transactions, the state machine for partial reconciliation. The thought is coherent, the sentences are flowing. Three minutes in, he hits a wall. His cloud voice tool cuts him off. Word limit reached. The thought is interrupted mid-sentence.

Next morning, he's piecing the fragmented notes back together, frustrated. Not at himself. At the tool. A meter interrupted him exactly when he was hitting clarity.

This is not rare. This is happening to developers across different roles and workflows.

The Bottleneck Actually Shifted

Marcus doesn't spend his day typing code anymore. Honestly, most developers don't. He spends it explaining bugs in Slack threads, drafting detailed PR reviews, writing design docs in Notion, refining prompts and specifications for Cursor and Claude. It's all long-form. It's faster than typing long-form. A word counter isn't a limit on speaking speed. It's a limit on thinking. That's everything when you're building with voice and LLMs.

The old voice tools were designed for a different problem: the typing problem. Developers type fast. Voice is slower. Voice adds friction. That was the frame in 2018. Today's reality is inverted. Developers are not typing code. They're typing intent. Intent is long-form. Intent is where the thinking lives.

LLM-driven workflows changed the shape of work. You're not writing loops. You're writing specifications. You're writing the context that a model will turn into loops, components, API handlers, documentation. Longer specifications. More detail. More thought. Fewer typos. Faster iteration. Voice is the native input layer for that workflow. A word meter is a hard ceiling on thinking.

The problem used to be: "Developers type code, they type fast, voice is slower than typing." That's solved. The new problem is: "Developers explain intent to models, they speak intent faster than they type it, and meters interrupt the explanation." Dictation breaking mid-thought is not a speed problem. It's a clarity problem.

Why Local Beats Cloud on Latency and Cost

Recitey runs Whisper locally on your device. Whisper-large-v3 achieves 96.3% word accuracy on LibriSpeech, production-grade transcription that runs on your own hardware. Free tier is uncapped. No meter. No variable cost per word. No network round trips waiting for results.

This distinction matters technically. Cloud-based dictation sends every spoken phrase to a vendor's servers. Network calls introduce latency. Latency accumulates: you stop speaking, your device makes a network call, waits for the cloud to process, gets the result back, and you can speak again. The rhythm fractures. You lose momentum. Your thought pauses. Local processing removes that loop. Dictation finishes the instant you stop speaking. No network wait. No timeout risk. No interruption.

The cost model compounds the advantage. Most voice tools charge per-word because they run on shared cloud infrastructure. Every transcription consumes CPU, storage, bandwidth, model inference. Metering makes economic sense for the vendor. Recitey inverts this: Whisper runs on your device, so there's zero variable cost. Your own hardware does the transcription work. Your own machine bears the CPU cost. Recitey's infrastructure cost per user for dictation is literally zero. Free tier has no limit because there's nothing to meter. The Pro tier charges for the rewrite pass, which is cloud-based cleanup, polish, and refinement. You're paying for actual compute value, not for the transcription you could run yourself.

Here's how the economics stack up:

The free tier is unlimited because there's nothing to limit.

Privacy and IP Security in Regulated Domains

Marcus works in fintech. Every Slack thread about payment logic touches regulated code. Every PR review on transaction processing contains company IP. Every postmortem on a production incident that touched customer funds is compliance-sensitive. He doesn't want those voice drafts uploaded to a vendor's cloud. He doesn't want retention policies and vendor practices becoming a compliance risk.

This extends beyond fintech. A healthcare engineer at a pharmacy software company dictating a bug fix for prescription-dispensing logic. A compliance officer at a bank reviewing code that handles customer funds. A data engineer at a SaaS platform writing specifications for billing systems. In each case, the spoken draft contains domain-specific IP, customer data references, or regulatory context that shouldn't leave the device.

HIPAA-regulated organizations can't use cloud transcription services without explicit business associate agreements. PCI compliance for payment systems restricts where cardholder data touches the network. GDPR restricts data flows outside the EU. SOC 2 compliance means thinking carefully about where customer data lives. Local processing sidesteps all of this. The voice never touches a third party's infrastructure.

With Recitey, Marcus's voice never leaves his laptop. The transcript stays on his device. If he chooses to send the refined version to Slack, to a doc, to email, that's his choice. But Recitey doesn't see the raw audio. Recitey doesn't transcribe it remotely. Recitey doesn't store it. Recitey doesn't build a retention policy around it.

The Pro tier adds a cloud rewrite pass if you want it, but that's optional. You can use Whisper locally for free and never touch the cloud. You can also choose to send specific drafts to the rewrite service when you want the polish. The choice is yours.

The Developer Personas This Actually Serves

Marcus is one shape of the problem. There are others.

Aisha is an API documentarian at a developer tools company. She dictates endpoint descriptions, request/response examples, error codes, usage guidelines. She works across six different documentation generators. She needs dictation that works everywhere, not locked into a single IDE or editor. Wispr Flow requires a Chrome extension. Otter.ai requires their app. Superwhisper requires Mac. Recitey works in whatever app she's already in: VS Code, Notion, Google Docs, terminal, even in-browser markdown editors. Free tier, no word limits, no platform lock-in.

Chen is a compliance engineer at a healthcare startup. He dictates audit notes, security procedure updates, incident response documentation. Every utterance might touch protected health information. Local processing is not a feature preference. It's a compliance requirement. A cloud vendor that stores audio transcripts is a liability.

Jordan is a technical writer documenting a REST API for internal use. She dictates the narrative explanations, the use cases, the warnings about pagination and rate limits. She's not working in a documentation CMS. She's drafting in a Google Doc, then moving it over. Cloud-only voice tools either don't work in her workflow or require special integrations.

Each of them hits the same friction: vendors that meter, vendors that require cloud, vendors that lock you into a particular tool or workspace. Recitey doesn't impose constraints. Free tier works across platforms, browsers, terminals. It stays local until you choose otherwise.

Why This Pricing Model Actually Makes Sense

The confusion around "free tier with no limits" usually boils down to a simple question: if dictation is free and unlimited, where's the revenue?

Recitey's answer is structural. Dictation is free because it has zero variable cost. Whisper runs on your machine. Your CPU. Your electricity. Your hardware. Recitey's infrastructure cost per user for dictation is zero.

The Pro tier is the rewrite pass. That's a cloud service. It's compute you don't own. It takes rough voice drafts, polishes them, ships back clean sentences in under 2 seconds. That's server cost. That's where you pay.

This is actually a cleaner model than most SaaS pricing. You pay for actual compute value, not for tiering artificial scarcity into your free tier. You're not hitting a word counter because the vendor wants to funnel you toward Pro. You're hitting nothing because there's nothing to hit.

And honestly, most of the time developers using the free tier don't actually need the Pro rewrite service. Whisper already produces clean text. The rewrite is for polish, not for fixing transcription errors. It's an optional enhancement, not a required step.

Real Friction Is Still Upstream

Marcus doesn't lose time to transcription latency anymore. His device handles that. His thought stays intact. His design doc flows.

The real constraint is upstream: how clearly he can articulate the problem, how deeply he can think through the solution on voice, how much context he can hold before speaking gets chaotic.

A word meter doesn't solve that problem. It makes it worse. It interrupts. It breaks the rhythm. It forces you to stop mid-thought and decide what to cut.

A tool that understands the shape of modern developer work, that stays on your device, that works everywhere, that doesn't interrupt, that doesn't meter, that serves the real problem.

More posts