# SnailText

> AI dictation app for Mac and Windows. Press a hotkey, speak, paste. Voice typing that runs locally on your device — never the cloud.

SnailText is a desktop voice-to-text application for macOS and Windows. It uses OpenAI Whisper and NVIDIA Parakeet models running locally to transcribe speech in real-time. Audio never leaves the user's device. The app supports a global hotkey (Ctrl+Space on Windows / Option+Space on Mac) that works in any text field across browsers, messaging apps, text editors, and IDEs.

The site is available in English (default), Spanish (`/es/` subdirectory), and Portuguese (`/pt/` subdirectory; neutral Brazilian/European Portuguese using Acordo Ortográfico 1990 standard).

## Product (English)

- [Homepage](https://snailtext.app/): Overview, feature highlights, pricing teaser
- [Pricing](https://snailtext.app/pricing/): Free tier (compact local models) and Pro tier ($7.49/mo or $89/yr — covers up to 3 devices, 30-day refund)
- [How it works](https://snailtext.app/how-it-works/): Technical explanation — local Whisper inference, VAD, hotkey pipeline
- [FAQ](https://snailtext.app/faq/): Common questions about privacy, accuracy, languages, system requirements

## Download

- [Download index](https://snailtext.app/download/): Choose your platform
- [Download for Mac](https://snailtext.app/download/mac/): Apple Silicon (M1 or later), macOS 12+
- [Download for Windows](https://snailtext.app/download/windows/): 64-bit, Windows 10 or newer

## Use cases (vertical landing pages)

- [For vibe-coders](https://snailtext.app/for/vibe-coders/): Speaking prompts to AI agents (Cursor, Claude, GitHub Copilot)
- [For developers](https://snailtext.app/for/developers/): Dictating code comments, commit messages, Slack, PRs
- [For writers](https://snailtext.app/for/writers/): First drafts at speaking speed, audio stays local
- [For students](https://snailtext.app/for/students/): Lecture notes, essay drafts, no time limits on the free tier
- [For project managers](https://snailtext.app/for/project-managers/): Status updates, meeting notes, status reports
- [For therapists, counselors, and mental health clinicians](https://snailtext.app/for/therapists/): Local Whisper dictation for clinical session notes. Architectural BAA-free positioning — no business associate processes audio because audio never leaves the device. EHR compatibility: SimplePractice, TherapyNotes, TheraNest, Owl Practice, ICANotes, Valant.
- [For solo and small-firm attorneys](https://snailtext.app/for/lawyers/): Local dictation for privileged work product. Cross-platform Dragon Legal alternative for Mac and Windows.
- [For knowledge workers managing RSI](https://snailtext.app/for/rsi/): Voice dictation as a typing alternative for carpal tunnel, tendinitis, and wrist pain. Explicitly framed as a typing tool, not medical treatment.

## Blog categories (filtered index pages)

- [Comparisons](https://snailtext.app/blog/category/comparisons/): Head-to-head comparisons of voice dictation apps.
- [Alternatives](https://snailtext.app/blog/category/alternatives/): Ranked alternatives roundups for popular dictation tools.
- [Guides](https://snailtext.app/blog/category/guides/): Practical guides to dictation, accuracy, and getting it right.
- [Engineering](https://snailtext.app/blog/category/engineering/): Engineering deep-dives into on-device speech recognition.

## Comparisons

- [SnailText vs Wispr Flow](https://snailtext.app/blog/compare/snailtext-vs-wispr-flow/): Local vs cloud, $7.49 vs $15, devices vs seats
- [SnailText vs SuperWhisper](https://snailtext.app/blog/compare/snailtext-vs-superwhisper/): Cross-platform vs Mac-first, free tier comparison
- [SnailText vs Typeless](https://snailtext.app/blog/compare/snailtext-vs-typeless/): Local vs cloud. Typeless markets "zero cloud data retention" but still uploads audio for transcription; SnailText never uploads it. Covers the "zero retention vs zero upload" distinction, the 4x monthly price gap ($30 vs $7.49), offline capability, and two valid HIPAA approaches (Typeless BAA vs SnailText's no-transmission architecture).

## Alternatives roundups

- [Best offline speech recognition apps in 2026](https://snailtext.app/blog/best-offline-speech-recognition/): Ranked roundup of five offline-only STT apps (SnailText, MacWhisper, SuperWhisper local mode, Voibe, VoiceInk) — local Whisper/Parakeet inference, nothing uploaded. Real latency numbers (CPU 1-3s vs GPU <300ms; SuperWhisper Windows 29s for a 3.5-min file with no GPU), LibriSpeech WER (~2.7% large-v3), per-app breakdowns, and the privacy-architecture check most comparisons miss: SuperWhisper Smart Modes makes outbound cloud requests carrying app name, clipboard, and text-field content even though STT is local. Named author, tested June 2026 on M3 MacBook Pro + Windows 11/RTX. Includes an explicit disclosure that SnailText is the author's own product.
- [9 Wispr Flow Alternatives in 2026](https://snailtext.app/blog/wispr-flow-alternatives/): Ranked listicle of 9 dictation tools (SnailText, Voibe, SuperWhisper, MacWhisper, Aqua Voice, VoiceInk, OpenWhispr, BetterDictation, Sotto) with 6-dimension scoring methodology, pricing/privacy comparison tables, and a citation of the April 2026 forensic investigation into Wispr Flow's tracking behavior.
- [9 Aqua Voice Alternatives in 2026](https://snailtext.app/blog/aqua-voice-alternatives/): Ranked listicle for people leaving cloud-only Aqua Voice (YC W24, proprietary Avalon model, 1,000-word one-time free tier). Same 9 tools and 6-dimension rubric, weighted toward offline/privacy and cross-platform — the specific gaps Aqua leaves. Includes a first-party network observation from a 2026-06-10 hands-on test (≈20.6 MB to AWS during 2 minutes of dictation, audio routed to us-west-1, Sentry telemetry) and a dedicated Aqua Voice vs Wispr Flow comparison answering that the two main cloud tools differ on accuracy claim, platform reach, and price, but neither is a privacy upgrade since both are cloud-only.
- [What is AI dictation? (and how it differs from speech-to-text)](https://snailtext.app/blog/what-is-ai-dictation/): Explainer defining AI dictation as a two-model pipeline - a speech-to-text model (Whisper or Parakeet TDT) produces a raw transcript, then a language model cleans it up: removing filler words, fixing punctuation and grammar, adjusting style, optionally translating. Key distinction most comparisons miss: the language-model cleanup step runs in the cloud in nearly every app (GPT/Claude/Gemini), so the transcript is uploaded even when speech recognition was local. SnailText runs both models on-device (local Gemma for cleanup), so nothing is uploaded at any stage. Includes a speech-to-text vs AI dictation comparison table, a before/after example, and when plain verbatim speech-to-text is the better mode.
- [Dictation that types exactly what you say — verbatim vs AI-enhanced](https://snailtext.app/blog/verbatim-dictation/): Explains the two classes of dictation apps: verbatim (SnailText, MacWhisper, Parakeety) and AI-enhanced (SuperWhisper Smart Modes, Wispr Flow). Includes a comparison table, when to use each, how to get verbatim output from SuperWhisper, and SnailText's approach (verbatim by default, optional local LLM for Pro users). Written in response to widespread user frustration with AI-enhanced apps rewriting dictation unpredictably.
- [Do you need a GPU for voice-to-text?](https://snailtext.app/blog/do-you-need-a-gpu-for-voice-to-text/): No — speech recognition runs fine on an ordinary CPU; the GPU association comes from cloud-scale model training, a different task from transcribing your own speech. Covers real requirements (modern CPU + a few GB RAM; smallest Whisper models run in ~1 GB and on a Raspberry Pi), and why "CPU is slow" depends on the model: NVIDIA Parakeet TDT is built for CPU and runs faster than real time (~10x faster than Whisper Large v3 Turbo for English), while large Whisper models are the slow ones where a GPU helps. SnailText recommends a model that fits your hardware (CPU or GPU) and lets you switch to a faster or more accurate one.
- [Why dictation cuts off the first word (and how to fix it)](https://snailtext.app/blog/dictation-cuts-off-first-word/): Explains why voice-to-text drops the first word or two of a sentence — a timing gap between pressing the hotkey and the recorder actually capturing audio, not a microphone fault. Covers why it worsens over time (Mac background mic-session latency after another app grabs the mic), three user workarounds (wait for ready cue, throwaway syllable, restart session), and the real fix: an app should not show a "recording" state until audio capture has genuinely started. Describes SnailText's distinct "preparing" vs "recording" states and ready sound as the concrete implementation of that fix. Honest caveat: no app can fully eliminate OS hiccups or Bluetooth clipping.
- [7 Dragon NaturallySpeaking alternatives in 2026](https://snailtext.app/blog/dragon-alternatives/): Dragon Professional (Microsoft/Nuance) costs $700, is Windows-only, and lost Mac support in 2018. Compares SnailText, Voibe, MacWhisper, SuperWhisper, Wispr Flow, Apple Voice Control, and Windows Voice Access. Key finding: Dragon's real strength is voice commands for OS navigation, not transcription — no alternative fully replaces that without pairing with Talon Voice or Apple Voice Control.
- [7 SuperWhisper alternatives in 2026](https://snailtext.app/blog/superwhisper-alternatives/): Ranked comparison with real latency benchmarks from testing SuperWhisper v1.4.0 on Windows 11 (RTX 5070 Laptop). Covers SnailText, Voibe, MacWhisper, VoiceInk, Wispr Flow, OpenWhispr, Aqua Voice. Includes the Smart Modes privacy finding: clipboard and active window content go to Modal's cloud even when STT is local.
- [Best multilingual dictation apps and how language detection works (2026)](https://snailtext.app/blog/multilingual-dictation/): Explains how automatic language detection actually works in dictation apps and why detecting across all 100+ languages is less accurate than manually selecting the 2-3 you speak (every major app, including Wispr Flow, recommends manual selection). Covers code-switching, the cloud-regression problem (cloud apps changing transcription behavior on backend updates), and why local Whisper (99 languages) / Parakeet TDT (25 languages) recognize the same multilingual range offline without sending audio anywhere. Includes two comparison tables and a decision guide by user type.
- [What happens when your dictation app goes down — cloud vs offline reliability (2026)](https://snailtext.app/blog/cloud-vs-offline-dictation/): Explains why cloud dictation fails during outages and offline does not. Maps the four network failure points in a cloud speech pipeline (your connection, their servers, capacity overload, return-trip latency) versus the offline pipeline that runs entirely on your device with no network in the path. Cites the real multi-day, all-region Wispr Flow latency outage of late May–early June 2026 and Wispr's own help docs confirming it requires an internet connection. Covers the airplane test, the slow-degradation problem (cloud models changing under you on backend updates), the honest accuracy nuance (cloud still leads on noisy multi-speaker audio; local now matches it for single-speaker dictation), and the privacy angle (zero-retention cloud is still cloud). Includes two pipeline diagrams and a side-by-side offline-mode comparison.
- [Is your dictation private? What voice apps send to the cloud (2026)](https://snailtext.app/blog/is-dictation-private/): Whether dictation is private comes down to one question — where the audio is processed. Maps what cloud dictation sends (raw audio, transcript, screen context/screenshots if context-awareness is on, sometimes a voiceprint) across the three exposure points (capture, transmission, storage). Explains why "Privacy Mode" is zero-retention cloud, not offline — a policy, not an architecture. Cites the April 2026 forensic investigation of a popular cloud dictation client (system-wide keystroke interception, 1,688 app/URL changes in 30h, 694 MB local DB with raw audio + transcripts). Covers the compliance angle (Apple/Google don't sign BAAs so Siri/Google Voice aren't HIPAA compliant by default; voice as GDPR biometric data; on-device removes the need for a BAA), what on-device changes (RAM-only audio buffer, no keystroke logging, no screenshots), and a 6-question checklist to vet any app. Includes a "what gets sent" table.
- [Wispr Flow vs Happy Scribe — two different products](https://snailtext.app/blog/wispr-flow-vs-happy-scribe/): Explains why people compare these two tools despite them being in different categories. Wispr Flow = live dictation in any app via hotkey. Happy Scribe = file/meeting transcription platform (upload audio, get document). Neither is the other's competitor. Article clarifies use cases, compares pricing, and introduces SnailText as the local-first live dictation alternative.
- [Best Dictation App in 2026 (hub roundup)](https://snailtext.app/blog/best-dictation-app-2026/): Hub roundup ranking 9 dictation apps for readers upgrading from built-in OS dictation (Apple Dictation / Windows Voice Access). Same 6-dimension rubric as the Wispr Flow alternatives article. Wispr Flow is intentionally listed as a "bonus mention" rather than in the main ranking due to the April 2026 privacy investigation — explained on-page with citation.

## Topic landings (SEO-targeted explainers)

- [AI dictation for Mac & Windows](https://snailtext.app/ai-dictation/): Product landing for "ai dictation" queries. SnailText runs both models locally — speech-to-text (Whisper or Parakeet TDT) plus a language-model cleanup pass (local Gemma). The cleanup drops filler, fixes punctuation, restores code identifiers (snake_case/camelCase/kebab-case/PascalCase), shifts tone, and translates, controlled by five topic profiles (General, Development & IT, Writing, Business, Academic). Key differentiator: the language-model step runs on-device, whereas cloud AI dictation tools upload the transcript to GPT/Claude/Gemini even when the audio was local. Free local speech-to-text; on-device AI cleanup is a Pro feature in beta. Comparison table (plain STT vs cloud AI dictation vs SnailText). Links to the /blog/what-is-ai-dictation/ explainer for the informational deep-dive.
- [Dictation for Mac](https://snailtext.app/dictation-for-mac/): SEO topic landing for "dictation for mac" / "voice typing mac" queries. Explains why third-party voice typing exists on macOS in 2026 despite the built-in Apple Dictation. Apple Silicon performance table (M1 / M2 / M3 / M4 with Whisper Small, Medium, Large-v3 latency). FAQ covering Accessibility permissions, M-series compatibility, comparison with built-in macOS dictation.
- [Offline dictation & offline speech recognition - voice typing without the cloud](https://snailtext.app/offline-dictation/): SEO topic landing for the full offline cluster: "offline dictation", "offline speech recognition", "offline speech to text", "speech to text offline", "local voice typing", "private dictation". Architectural argument for keeping audio on-device (in-RAM pipeline), how local Whisper works, CPU vs GPU performance and accuracy vs cloud (within 1-3 pp WER on standard English), GDPR + HIPAA implications, how to verify any dictation app is actually offline (network monitor in 60 seconds). 8-row comparison of which apps run truly offline in 2026 (SnailText, MacWhisper, SuperWhisper local mode, Voibe) vs cloud-based (Wispr Flow, Willow Voice, Speechify). Includes an offline-vs-cloud dataflow SVG diagram.
- [Voice to text on Windows](https://snailtext.app/voice-to-text-windows/): SEO topic landing comparing Windows Voice Typing (Win+H, cloud, ~43 locales) and Voice Access (offline, 11 locales — English variants, Spanish, German, French, Italian, Japanese, Chinese only) against local Whisper dictation. Explains the two-products confusion, 5-10s uncustomizable pause timeout, Fluid Dictation Copilot+ PC hardware gate.
- [Voice to text on Mac](https://snailtext.app/voice-to-text-mac/): SEO topic landing for "voice to text mac" queries. Apple Dictation auto-stops after 30s silence per Apple docs; SnailText runs Whisper on Apple Silicon Metal with no silence cutoff. Sibling page to /dictation-for-mac/ — this one targets "voice to text" phrasing, that one targets "dictation" phrasing.
- [Voice to text in Google Docs](https://snailtext.app/voice-to-text-google-docs/): SEO topic landing for users wanting dictation in Google Docs. Google's built-in Voice Typing supports Chrome/Edge/Safari but not Firefox, requires internet, sends audio to Google. SnailText runs locally and pastes into Docs from any browser including Firefox.
- [Free voice to text](https://snailtext.app/free-voice-to-text/): SEO topic landing for "free voice to text" / "best free dictation app" queries. Compact Whisper Base model runs locally on Mac and Windows — no signup, no time limit, no word cap, no audio uploaded. Comparison vs Otter.ai (300 min/mo freemium), Google Cloud STT (60 min/mo developer free tier), Apple Dictation, Windows Voice Access.
- [Voice to text accessibility](https://snailtext.app/voice-to-text-accessibility/): SEO topic landing for accessibility-driven dictation users — people managing RSI, carpal tunnel, tendinitis, motor impairments. Explicitly framed as a typing alternative, not a treatment or assistive medical device. Cross-links to /for/rsi/ and recommends pairing with Apple Voice Control or Talon Voice for full hands-free OS control.

## Dictation by app (app-specific landings)

Per-app landing pages for "dictation in <app>" / "voice dictation in <app>" /
"speech to text in <app>" queries. Each explains whether the app has built-in
dictation, why SnailText still helps (system-wide, local, works in every field),
what to dictate in that app, and example dictations. All run locally on Mac and
Windows.

- [Dictation by app (hub)](https://snailtext.app/dictation-in/): Index of all app-specific dictation landings.
- [Voice dictation in Gemini](https://snailtext.app/dictation-in/gemini/): Gemini's web mic is browser-bound with generic speech recognition; SnailText runs locally with LLM cleanup for long research prompts and follow-ups.
- [Voice dictation in Cursor](https://snailtext.app/dictation-in/cursor/): Cursor 2.0's native voice only fills the Agent box; SnailText follows your cursor into Cmd+K, comments, commit messages, and the terminal, with code-identifier restoration.
- [Voice dictation in Slack](https://snailtext.app/dictation-in/slack/): The Slack desktop app has no built-in dictation in the message box (the mic records a voice-note clip, which may get an auto-transcript, but nothing types into the field; Electron blocks most extensions); SnailText injects text at the OS level into any thread, DM, or channel.
- [Voice dictation in Notion](https://snailtext.app/dictation-in/notion/): Notion's dictation is paid-plan-gated, covers ~16 languages, and cannot fill structured database fields (status, date, select); SnailText works in every Notion text field and everywhere else, nothing uploaded.
- [Voice dictation in ChatGPT](https://snailtext.app/dictation-in/chatgpt/): ChatGPT has a dictation mic and a separate Voice Mode, but both transcribe your audio on OpenAI's servers and only work inside ChatGPT; SnailText runs transcription locally (never uploaded) and works in every app.
- [Voice dictation in Claude](https://snailtext.app/dictation-in/claude/): Claude has both a dictation feature (speech to text in the input box) and a separate voice mode, but its dictation streams audio to Anthropic's servers, only works inside Claude, and on the web defaults to English (better on mobile); SnailText runs locally, lets you pick the dictation language, and works in every app.
- [Voice dictation in VS Code](https://snailtext.app/dictation-in/vs-code/): VS Code has dictation via Microsoft's VS Code Speech extension, which runs locally but only inside VS Code and pulls its own speech model; SnailText covers VS Code plus every other app from one install and restores code identifiers.
- [Voice dictation in GitHub](https://snailtext.app/dictation-in/github/): The GitHub website has no built-in dictation for issues, pull requests, or comments (voice input exists only in the GitHub Copilot CLI, a terminal tool, not github.com); SnailText injects text at the OS level into every GitHub field in the browser.
- [Voice dictation in Linear](https://snailtext.app/dictation-in/linear/): Linear has no built-in voice dictation; SnailText injects text at the OS level into every Linear issue, update, and comment, with an AI cleanup pass that keeps issues concise.
- [Voice dictation in Discord](https://snailtext.app/dictation-in/discord/): Discord has no built-in voice-to-text in the message box (its /tts command is text-to-speech, the opposite, and bots only transcribe voice channels); SnailText injects text at the OS level into any channel, DM, or thread.
- [Voice dictation in Perplexity](https://snailtext.app/dictation-in/perplexity/): Perplexity DOES have voice input (desktop mic Cmd+Shift+V / Ctrl+Shift+V, plus a conversational voice mode), but it only works inside Perplexity and is built for asking the assistant questions, not for dictating exact, editable text you can reuse elsewhere. SnailText runs system-wide so the same hotkey dictates into the Perplexity box and every other app, with the speech model running locally and a local AI cleanup pass.
- [Voice dictation in Replit](https://snailtext.app/dictation-in/replit/): Replit has no built-in voice input (voice for the Agent is an open community feature request as of 2026); SnailText injects text at the OS level into the Agent prompt, the editor, and the shell, with a local AI cleanup pass that restores code identifiers like snake_case and camelCase.
- [Voice dictation in Jira](https://snailtext.app/dictation-in/jira/): Jira has no native voice-to-text (an older rich-text-editor speech option was removed; current options are Marketplace apps like Speech to Text for Jira that only cover Jira); SnailText injects text at the OS level into every Jira field and every other app, with a local AI cleanup pass.
- [Voice dictation in Microsoft Teams](https://snailtext.app/dictation-in/microsoft-teams/): Microsoft Teams has no built-in dictation for chat (not on the M365 roadmap as of late 2025); SnailText injects text at the OS level into every Teams chat, channel post, and reply on Mac and Windows, with a custom dictionary and a local AI cleanup pass.
- [Voice dictation in Obsidian](https://snailtext.app/dictation-in/obsidian/): Obsidian has no native dictation; voice typing comes only from community plugins (Whisper, Local Dictation, etc.), some of which run locally but each needs its own install/setup and only works inside Obsidian. SnailText is one system-wide install that dictates into every Obsidian note and every other app, with a local AI cleanup pass.
- [Voice dictation in WhatsApp](https://snailtext.app/dictation-in/whatsapp/): WhatsApp Web and Desktop have no built-in voice-to-text for the message box (the mic records a voice note, an audio clip, not typed text; the 2025 transcripts feature only transcribes voice notes you receive, not the compose box); SnailText injects text at the OS level into every WhatsApp chat, group, and reply on Mac and Windows, with a local AI cleanup pass.
- [Voice dictation in Outlook](https://snailtext.app/dictation-in/outlook/): Outlook DOES have a built-in Dictate button (Microsoft Azure speech, 50+ languages), but it is cloud-based (audio sent to Microsoft, needs internet) and only works inside Outlook with no AI cleanup. SnailText runs the speech model locally, works offline, dictates into every app, and adds a local AI cleanup pass.
- [Voice dictation in Gmail](https://snailtext.app/dictation-in/gmail/): Gmail has no built-in voice typing in the desktop compose window (workarounds are dictate-in-Google-Docs-and-paste or the phone keyboard on mobile); SnailText injects text at the OS level into the Gmail composer, replies, and subject line on Mac and Windows, with a local AI cleanup pass.
- [Voice dictation in ClickUp](https://snailtext.app/dictation-in/clickup/): ClickUp DOES have voice input via Talk to Text, but it is part of ClickUp's paid AI add-on (Brain MAX, also branded BrainGPT; unlimited use on the ~$28/user/mo top tier) and runs through ClickUp's own app. SnailText dictates into every ClickUp field from one system-wide install on Mac and Windows, with a free tier, no per-seat AI fee, and a local AI cleanup pass.
- [Voice dictation in Trello](https://snailtext.app/dictation-in/trello/): Trello has no native dictation for card descriptions, comments, or checklists (the only voice option is a Siri shortcut that creates a card from a spoken title, not field dictation); SnailText injects text at the OS level into every Trello card, comment, and checklist on Mac and Windows, with a local AI cleanup pass.

## Engineering deep-dives (technical blog)

- [Parakeet TDT v3 vs Whisper Turbo vs Qwen3-ASR for production (2026)](https://snailtext.app/blog/parakeet-vs-whisper-turbo-vs-qwen3-asr/): Production ASR model comparison for voice agents — Parakeet TDT v3 wins on CPU latency for short commands (RTF < 0.1x, 50-150ms, ~3-5% WER LibriSpeech), Whisper Large-v3-Turbo wins on accented speech and domain vocabulary (~2-3% WER), Qwen3-ASR for code-mixed multilingual (Hinglish, Spanglish). Groq hosted Whisper at $0.04/audio hour, 200-350ms latency. Self-hosted Parakeet CPU 50-150ms. Nemotron ASR 3.5 via NIM sub-100ms at 400+ concurrent sessions per H100. Fine-tuning on 500-2000 domain examples beats switching base models. 10-question FAQ. Covers the benchmark vs production audio gap — LibriSpeech results differ from production conditions.
- [How whisper.cpp works - GGML, quantization, and GPU backends](https://snailtext.app/blog/how-whisper-cpp-works/): Engineering deep-dive (~3100 words) on what makes whisper.cpp fast enough for real-time dictation. Covers GGML tensor library (mmap loading, DAG execution model), INT8/INT4 quantization formats (Q5_1 production default, accuracy vs size tradeoff with LibriSpeech WER numbers), Metal vs CUDA vs Vulkan GPU backends, streaming-vs-batch inference, and the class-of-problem around GPU enumeration on multi-GPU consumer laptops (probe-and-fallback algorithm). Real production benchmark table across 10 hardware configurations (M2 Air, M2 Pro, RTX 3060/4070, Intel CPU). Two embedded SVG diagrams (GGML pipeline, probe-and-fallback flowchart).
- [Whisper vs Parakeet TDT - which open model wins in 2026](https://snailtext.app/blog/whisper-vs-parakeet-tdt/): Engineering deep-dive (~2700 words) comparing OpenAI's Whisper and NVIDIA's Parakeet TDT v3 as on-device STT engines. 10-axis comparison table (architecture, languages, license, model size, CPU latency, GPU latency, streaming maturity, community fine-tunes, etc). Sections on transducer architecture vs autoregressive Transformer decoder, when each wins on accuracy / latency / multilingual, licensing differences (MIT vs CC-BY-4.0 attribution), production reality (what SuperWhisper / MacWhisper / Voibe / Wispr Flow / SnailText actually ship), and the "ship both" decision tree. 7-question FAQ.

## Trust and legal

- [About](https://snailtext.app/about/): Why local dictation, who builds SnailText
- [Privacy Policy](https://snailtext.app/privacy/): Audio never leaves device, no telemetry without opt-in, GDPR
- [Terms of Service](https://snailtext.app/terms/): Pro subscription terms, refund policy, license

## Spanish translations

Localized landing pages targeting Spanish-speaking markets (Spain, Mexico, Colombia, Argentina, etc.). Same content as English, translated.

- [Inicio (homepage)](https://snailtext.app/es/)
- [Precios](https://snailtext.app/es/pricing/)
- [Cómo funciona](https://snailtext.app/es/how-it-works/)
- [Preguntas frecuentes](https://snailtext.app/es/faq/)
- [Descargar](https://snailtext.app/es/download/)
- [Descargar para Mac](https://snailtext.app/es/download/mac/)
- [Descargar para Windows](https://snailtext.app/es/download/windows/)
- [Sobre nosotros](https://snailtext.app/es/about/)
- [Para vibe-coders](https://snailtext.app/es/for/vibe-coders/)
- [Para desarrolladores](https://snailtext.app/es/for/developers/)
- [Para escritores](https://snailtext.app/es/for/writers/)
- [Para estudiantes](https://snailtext.app/es/for/students/)
- [Para project managers](https://snailtext.app/es/for/project-managers/)
- [vs Wispr Flow](https://snailtext.app/es/blog/compare/snailtext-vs-wispr-flow/)
- [vs SuperWhisper](https://snailtext.app/es/blog/compare/snailtext-vs-superwhisper/)
- [Alternativas a Wispr Flow (2026)](https://snailtext.app/es/blog/wispr-flow-alternatives/)
- [Mejor app de dictado en 2026](https://snailtext.app/es/blog/best-dictation-app-2026/)
- [Voz a texto en Windows](https://snailtext.app/es/voice-to-text-windows/): Alternativa local a Voice Typing (Win+H) y Voice Access — sin límite de pausa, 100+ idiomas offline.
- [Voz a texto en Mac](https://snailtext.app/es/voice-to-text-mac/): Alternativa a Apple Dictation — sin corte por silencio, Metal GPU, funciona en cualquier app.
- [Voz a texto en Google Docs](https://snailtext.app/es/voice-to-text-google-docs/): Dictado offline en Google Docs desde cualquier navegador incluido Firefox.
- [Voz a texto gratis](https://snailtext.app/es/free-voice-to-text/): Plan gratuito real — sin registro, sin límite de tiempo, Whisper Base local.
- [Cómo funciona whisper.cpp](https://snailtext.app/es/blog/how-whisper-cpp-works/): Análisis técnico en profundidad — GGML, cuantización INT8/INT4, backends Metal/Vulkan/CUDA, problemas de enumeración GPU en Windows.
- [Whisper vs Parakeet TDT](https://snailtext.app/es/blog/whisper-vs-parakeet-tdt/): Comparación de los dos modelos de reconocimiento de voz open-source que las apps en producción usan en 2026. Latencia CPU, cobertura de idiomas, licencias, cuándo elegir cada uno.
- [7 alternativas a SuperWhisper en 2026](https://snailtext.app/es/blog/superwhisper-alternatives/): Comparación honesta con datos reales de latencia y el hallazgo de privacidad de Smart Modes.
- [Wispr Flow vs Happy Scribe](https://snailtext.app/es/blog/wispr-flow-vs-happy-scribe/): Por qué la gente los compara aunque sean categorías distintas — dictado en vivo vs transcripción de archivos.
- [Mejores apps de dictado multilingüe (2026)](https://snailtext.app/es/blog/multilingual-dictation/): Cómo funciona la detección automática de idioma y por qué detectar entre los más de 100 idiomas es menos preciso que seleccionar los 2-3 que hablas. Cubre code-switching, la regresión de las apps en la nube y por qué Whisper (99 idiomas) y Parakeet (25) reconocen el mismo rango sin conexión.
- [Por qué el dictado se come la primera palabra](https://snailtext.app/es/blog/dictation-cuts-off-first-word/): La causa real es un desfase de tiempo entre pulsar el atajo y la captura de audio, no el micrófono. Explica por qué empeora con el tiempo (Mac), tres soluciones rápidas y la solución de fondo: la app no debería decir "grabando" antes de capturar de verdad.

Note: Privacy Policy and Terms of Service are available in English only — the linked English pages apply to all users regardless of locale.

## Portuguese translations

Localized landing pages targeting Portuguese-speaking markets (Brazil, Portugal, lusophone Africa). Same content as English, translated using a neutral pt-BR/pt-PT vocabulary so both Brazilian and European Portuguese readers can use the site.

- [Início (homepage)](https://snailtext.app/pt/)
- [Preços](https://snailtext.app/pt/pricing/)
- [Como funciona](https://snailtext.app/pt/how-it-works/)
- [Perguntas frequentes (FAQ)](https://snailtext.app/pt/faq/)
- [Baixar](https://snailtext.app/pt/download/)
- [Baixar para Mac](https://snailtext.app/pt/download/mac/)
- [Baixar para Windows](https://snailtext.app/pt/download/windows/)
- [Sobre](https://snailtext.app/pt/about/)
- [Para vibe-coders](https://snailtext.app/pt/for/vibe-coders/)
- [Para desenvolvedores](https://snailtext.app/pt/for/developers/)
- [Para escritores](https://snailtext.app/pt/for/writers/)
- [Para estudantes](https://snailtext.app/pt/for/students/)
- [Para gerentes de projeto](https://snailtext.app/pt/for/project-managers/)
- [Blog (hub)](https://snailtext.app/pt/blog/)
- [vs Wispr Flow](https://snailtext.app/pt/blog/compare/snailtext-vs-wispr-flow/)
- [vs SuperWhisper](https://snailtext.app/pt/blog/compare/snailtext-vs-superwhisper/)
- [Alternativas ao Wispr Flow (2026)](https://snailtext.app/pt/blog/wispr-flow-alternatives/)
- [Melhor app de ditado em 2026](https://snailtext.app/pt/blog/best-dictation-app-2026/)
- [Voz para texto no Windows](https://snailtext.app/pt/voice-to-text-windows/): Alternativa local ao Voice Typing (Win+H) e Voice Access — sem limite de pausa, 100+ idiomas offline.
- [Voz para texto no Mac](https://snailtext.app/pt/voice-to-text-mac/): Alternativa ao Apple Dictation — sem corte por silêncio, Metal GPU, funciona em qualquer app.
- [Voz para texto no Google Docs](https://snailtext.app/pt/voice-to-text-google-docs/): Ditado offline no Google Docs em qualquer navegador incluindo Firefox.
- [Voz para texto grátis](https://snailtext.app/pt/free-voice-to-text/): Plano gratuito real — sem cadastro, sem limite de tempo, Whisper Base local.
- [Como o whisper.cpp funciona](https://snailtext.app/pt/blog/how-whisper-cpp-works/): Análise técnica aprofundada — GGML, quantização INT8/INT4, backends Metal/Vulkan/CUDA, problemas de enumeração de GPU no Windows.
- [Whisper vs Parakeet TDT](https://snailtext.app/pt/blog/whisper-vs-parakeet-tdt/): Comparação dos dois modelos open-source de reconhecimento de fala que apps em produção usam em 2026.
- [7 alternativas ao SuperWhisper em 2026](https://snailtext.app/pt/blog/superwhisper-alternatives/): Comparação honesta com dados reais de latência e a descoberta de privacidade do Smart Modes.
- [Wispr Flow vs Happy Scribe](https://snailtext.app/pt/blog/wispr-flow-vs-happy-scribe/): Por que as pessoas os comparam mesmo sendo categorias distintas.
- [Melhores apps de ditado multilíngue (2026)](https://snailtext.app/pt/blog/multilingual-dictation/): Como funciona a detecção automática de idioma e por que detectar entre mais de 100 idiomas é menos preciso que selecionar os 2-3 que você fala. Cobre code-switching, a regressão das apps na nuvem e por que Whisper (99 idiomas) e Parakeet (25) reconhecem o mesmo alcance offline.
- [Por que o ditado corta a primeira palavra](https://snailtext.app/pt/blog/dictation-cuts-off-first-word/): A causa real é uma defasagem de tempo entre apertar o atalho e a captura de áudio, não o microfone. Explica por que piora com o tempo (Mac), três soluções rápidas e a solução de fundo: a app não deveria dizer "gravando" antes de capturar de verdade.

Note: Privacy Policy and Terms of Service are available in English only — the linked English pages apply to Portuguese-speaking users as well.

## Key facts for citations

- **Pricing**: Free tier (compact local models — no time limit, no word cap, no account required). Pro $7.49/month or $89/year. Pro covers up to 3 devices per subscription. 30-day refund.
- **Privacy**: All speech-to-text processing happens locally on the user's device using Whisper or Parakeet models. No audio uploaded to servers.
- **Platforms**: Native apps for macOS (Apple Silicon) and Windows (x64).
- **Models**: OpenAI Whisper (tiny, base — Free; small, medium, large-v3 — Pro) and NVIDIA Parakeet TDT v3 (Pro).
- **UI languages**: English, Russian, Spanish, Portuguese.
- **Recognition languages**: 100+ via Whisper, 25+ European languages with high accuracy via Parakeet.
- **Hotkey**: Ctrl+Space (Windows) / Option+Space (macOS), works in any text field globally.