Dictation App Features Compared: Offline, Privacy, AI

Kept, 2 figuresweb:https://www.typeless.com

Home page says Typeless works across all apps and turns speech into polished text; it gives no offline statement, so no offline row.

Kept, 3 figuresweb:https://www.typeless.com/privacy

Privacy policy says voice audio is processed on Typeless's cloud servers (speech not on the device), that no recordings are stored on its servers, and that audio and text may go to third-party AI.

Not keptweb:https://www.typeless.com/pricing

Plan table lists 'On device' only beside 'On-device history storage', which is not a statement that speech is processed on the device; no other feature statement for these groups.

Not keptweb:https://handy.computer

Says the voice stays on the computer and that text goes into any text field; both are already rows from the project's GitHub README, so no new figure.

Kept, 2 figuresweb:https://handy.computer/privacy

Privacy policy describes optional AI post-processing through user-chosen providers and states that cloud post-processing sends text only when the user invokes it.

Not keptweb:https://handy.computer/docs

Repeats 'All models run entirely on your device' and 'Click into any text field', which the project's GitHub README already gives as rows; no new figure.

Kept, 3 figuresweb:https://support.microsoft.com/en-us/windows/use-voice-typing-to-talk-instead-of-type-on-your-pc-fec94565-c4bd-329d-e59a-af033fa5689f

Microsoft's help page says voice typing needs internet and uses online recognition (no offline, no on-device speech), and fluid dictation runs on on-device language models on Copilot+ PCs.

Kept, 1 figureweb:https://www.getvoibe.com

Home page headline says the voice is typed into any app, on Mac and Windows.

Kept, 3 figuresweb:https://www.getvoibe.com/cloud-ai-privacy

Explains two modes: On-Device Mode works with no internet and transcribes on the Mac; Cloud Mode uses cloud speech and an AI model that polishes the text.

Kept, 10 figuresWispr Flow

Wispr's own help center, pricing, privacy and home pages give a yes or no for all ten features: no offline transcription (help article), audio sent to its servers, AI polish, dictionary on the free plan, Command Mode for voice editing, opt-out training setting, optional cloud audio storage and third-party AI providers in its privacy policy.

Kept, 10 figuresSuperwhisper

Superwhisper's own home, models, docs, terms and billing pages give a yes or no for all ten features: works offline, on-device transcription with audio that never leaves the device, Super Mode rewrites and commands on selected text, free local Whisper plan with custom vocabulary, cloud models through third-party AI APIs, and terms that say data is not used for training and cloud audio is not retai

Kept, 6 figuresAqua Voice

Aqua Voice's own Windows, home and pricing pages give a yes or no for six features: it needs an internet connection and processes audio in the cloud, types into any Windows app, Avalon refines text with AI, custom dictionary, and a 1,000-word free plan. No row for voice editing (its Max plan 'Send it' command submits, it does not edit), for recordings training its models (Privacy Mode text says tr

Kept, 9 figuresVoiceInk

Official README, website, privacy policy and Modes docs give a figure for 9 of 10 features; no statement says recordings train the maker's models, so no row.

Kept, 8 figuresWillow Voice

Official pricing, security and privacy pages give 8 rows (one reused); no speech-on-device row because the page says dictation runs in the cloud and also says offline when needed, without detail; no voice-editing commands stated.

Kept, 8 figuresSpokenly

Official pricing, home, privacy, custom dictionary and cleanup docs give a figure for 8 features; the page names no voice commands for editing and no use of recordings for training, so no rows for those.

Kept, 8 figuresAmical by Amical

Amical's homepage, docs, FAQ, pricing and mobile privacy policy give rows for works offline, speech on the device, types into any app, AI formatting, custom vocabulary, a free plan, no model training, and no audio stored (mobile app). The homepage's voice commands are not described as editing commands, and no third-party AI provider is named for its cloud models, so those cells are left out.

Kept, 6 figuresAudionotes by Audionotes

Audionotes is a cloud note-taker. Its privacy policy names Deepgram and Assembly AI for speech-to-text, so speech is not processed on the device, and it names OpenAI and Google Gemini for text processing. The homepage and pricing give AI structuring, a free plan with 1-minute voice notes, no LLM training, and recordings stored in the account (the page does not use the word servers). No page mentio

Kept, 10 figuresBiBi Keyboard by BryceWG

The maker's homepage (Chinese) and docs give rows for offline recognition with local models, the floating ball that types into any keyboard, LLM rewriting, a Pro-only hotword list, voice-instruction editing, a free open-source edition, no personal data collection with no retained audio copy, and text or audio sent to a chosen cloud provider only when cloud ASR or AI post-processing is on. Pro pric

Kept, 6 figuresBossAI by Hyathi Technologies Private Limited

BossAI's own home and pricing pages give a yes or no for six features: a free plan with 200 words a day, Rewrite as AI, a 10-word free dictionary, Boss Mode voice commands, typing into all apps on macOS, and no data stored at its servers. The pages do not say the app works offline, does not name a third-party AI, and do not say whether recordings train models, so those rows are left out.

Kept, 6 figuresCapsWriter-Offline by HaujetZhao

The CapsWriter-Offline README (maker's GitHub page) gives a yes or no for six features: fully offline, on-device recognition, typing into programs on Windows, LLM roles, hot-word replacement, and audio saved as local files. The README does not state a license-based free plan, voice editing commands, or a third-party AI provider, so those rows are left out.

Kept, 2 figuresChirp by Whamp

The Chirp README (maker's GitHub page) gives a yes or no for two features: it runs fully locally on a Parakeet speech model, and it has configurable word overrides. It does not say the app works offline after setup, does not mention audio storage or AI post-processing, and its MIT license is not a plan, so those rows are left out.

Kept, 6 figuresCleft by Cleft

Its own pricing and privacy pages state offline use, on-device Whisper transcription, AI-structured notes, a $0 Basic plan, note-only text sent to OpenAI or Groq, and no AI training on user data. No row for typing into any app, custom vocabulary or voice editing commands: the pages do not say. No row for keeping audio on servers: the privacy policy both says audio is never sent to a server and say

Kept, 7 figuresClovy by Open Software Network

Its own product page and privacy policy state a $0 Free tier, dictation into any app, AI dictation cleanup, routing to Venice AI, that transcription audio goes to its backend, that recordings stay on the device by default, and that it does not train on user content. No row for offline use or on-device speech: the pages do not say dictation works without the network, and say transcription audio goe

Kept, 8 figuresCodictate by Emil Lykke Grann

Its own homepage and privacy policy state offline use after the model download, on-device speech models, pasting into any app, custom vocabulary, a free and open source app, no collection of audio recordings, no data sent to any server, and no data sent to OpenAI or NVIDIA. No row for AI rewriting: the homepage says Dictionary Mode writes what you meant but never says AI. No row for voice editing

Kept, 8 figuresContextli by Ertiqah, LLC.

Contextli's own pricing and privacy pages and home page give a yes or no for eight features: offline modes on paid plans, on-device audio in Local Whisper Mode, any app you can type in, AI formatting, no free plan (a free trial with a card), no training on recordings, no audio stored after transcription, and audio and text shared with Deepgram and xAI. The pages do not state custom vocabulary or v

Kept, 9 figuresDictaFlow by SmartBids.ai Corp.

DictaFlow's home page and privacy policy give a yes or no for nine features: offline dictation on Pro, speech on the device, typing into supported text fields (not every app), optional AI cleanup, custom vocabulary, desktop voice edits, a 2,000-word free plan, audio not kept permanently, and named AI transcription providers. The privacy policy does not say whether recordings train models, so that

Kept, 4 figuresDictanote by Dictanote

Dictanote's pricing page gives a yes or no for four features: notes that work offline (dictation offline not stated), a free plan with 60 minutes of dictation a day, edit by voice on Pro, and no audio received or stored by the company. The pages do not say whether speech is processed on the device, whether text goes to a third-party AI, whether recordings train models, or whether dictation works i

Kept, 8 figuresDictate Keyboard by DevEmperor

The developer's README and the official site give yes or no for eight features: offline dictation with a downloaded model, speech kept on the phone, dictation into any app, AI rewording, a personal dictionary, a freeform AI voice command on selected text, free and open source, and rewording sent to OpenAI, Anthropic or Gemini. The site's FAQ says cloud audio does not go through a developer server,

Kept, 5 figuresDictation Daddy by Dictation Daddy

The home, pricing and privacy pages give yes or no for five features: dictation into any app with one key, AI Cleanup and Enhancement, Templates and Custom Vocabulary, no training on audio, and audio stored on the maker's servers by default. The privacy policy conflicts with the site's FAQ, which says no audio is stored, and the policy is quoted. No statement on offline use, on-device speech or ed

Not keptDictation for Gmail by Dictation Tools

The official site at dictation.tools is a one-page landing page that links only to the Chrome Web Store listing and states no feature. The privacy policy never mentions audio, voice, speech, recordings or transcription, so it gives no figure for any dataset feature. The store listing was not used because it is not one of the official pages named for this target.

Kept, 8 figuresDictato by Dictato

Official home page states offline use, on-device speech, a 7-day trial then one-time 19.99 euro price, and that nothing is uploaded; no statement on voice editing commands or model training

Kept, 6 figuresDun by Manuel Fessen

Official privacy policy and home page state on-device transcription, a personal dictionary, optional on-device AI polish, and no voice upload or training; no statement on offline use, a free plan, or voice editing commands

Kept, 8 figuresEnviousWispr by EnviousWispr

Official home page states local speech engines, offline AI polish, a free app with no account, and optional own-key cloud polish; no statement on voice editing commands or model training

Kept, 5 figuresFlint by Flint

Official home page and privacy policy state on-device mode with no audio or text leaving the device, AI summaries, a free on-device tier, and cloud providers Deepgram, OpenAI and Anthropic; no statement on offline use, typing into other apps, custom words, voice editing commands or model training

Kept, 8 figuresFlow Keyboard by Snowline LLC

Official pricing, privacy and home pages state that voice features need internet, audio is sent to servers and not kept after processing, AI polishing, a free plan of 5,000 words a week, voice editing of selected text, system-wide typing on Mac and Windows, and unnamed cloud AI subprocessors; its home page structured data also lists a custom dictionary, but that text is not visible on the page, so

Kept, 7 figuresFluidVoice by Altic

Official home page and FAQ state local speech models that work offline on Mac, on-device transcription, typing into other apps with accessibility permission, on-device text cleanup, a free and open-source app, audio kept on the device by default, and optional cloud providers; no statement on custom vocabulary, voice commands for editing (Command Mode is described only as taking actions on the Mac)

Kept, 10 figuresFnScribe by FnScribe

Its own home and privacy pages state ten rows: works offline after a one-time model download, speech stays on the Mac, types into any Mac app, AI cleanup, word list, voice commands, free, audio never reaches the maker, and text to an optional AI provider.

Kept, 8 figuresFreestyle by Freestyle

Its pricing, home and privacy pages state eight rows: any desktop app, local models for speech, AI cleanup, Vocabulary and Dictionary, a Free plan, no training on content, cloud audio not kept, and OpenRouter. No row for works offline, since the pages say local transcription runs on the machine but never say the app works offline; no row for voice commands for editing.

Kept, 6 figuresFUTO Keyboard by FUTO

Its home, privacy and docs pages state six rows: 100% offline, offline voice input with no internet, a personal dictionary, no network requests by the app (audio not on servers, no text to AI), and voice recordings never shared. No Free plan row: the page prices the app at $6.99 or $11.99 but also says to pay if you like it, so the page does not settle it. No row for AI rewriting, voice editing co

Kept, 5 figuresGhost Pepper by matthartman

The GitHub README, the maker's own repository, gives 1 or 0 for five features: speech runs on the Mac with nothing sent anywhere (1), recordings stay on the Mac (0), it types into any text field (1), local LLM cleanup (1), and it is free and open source (1). No line says it works offline or adds custom words or voice editing, and the optional cloud features that need your own API key are not count

Kept, 7 figuresGlaido by Glaido

The home, pricing, privacy and Commands docs pages give 1 or 0 for seven features: a free tier, dictation into any app, optional AI formatting, a dictionary, spoken commands that rewrite selected text, and AI processing by Groq (all 1), and speech on the device (0, inferred from the privacy policy's statement that transcription runs at service providers). The pages do not say it works offline, and

Kept, 9 figuresGlimpse by Glimpse

The home and privacy pages give 1 or 0 for nine features: works with the wifi off, speech on the user's computer, types into any app, paid AI cleanup, spoken edits in place, a dictionary, free dictation, and an optional LLM with the user's own key (all 1), and audio never reaching its servers (0). The pages do not say whether recordings train the maker's models, so that row is left out.

Kept, 7 figuresGoogle AI Edge Eloquent by Google

The App Store page, written by developer Google, states offline use, on-device processing, no usage limits, an editable personal dictionary, AI rewriting, the iOS keyboard as coming soon, and audio staying on the device except for optional cloud features. No statement about model training or third-party AI.

Kept, 4 figuresHex by Kit Langton

The official site hex.kitlangton.com states only the version, platform and download, so its feature rows come from the anomalyco/hex README, which the site links as its GitHub download and the archived kitlangton/Hex repo names as the new source. The README gives on-device transcription, paste into the active app, optional post-processing to a model provider, and Modes corrections; it does not say

Kept, 6 figureshns by primaprashant

The documentation and homepage state offline use after the model download, local Whisper processing with no data leaving the device, clipboard copy rather than typing, no subscription, and no use of data for model training. Rewriting and third-party AI are only user-piped LLM examples and a custom vocabulary is not mentioned, so no rows for those.

Kept, 7 figuresMonologue

Maker's home page and privacy policy give seven features: offline local models, Mac-only local transcription, any-app text, personal dictionary, no dictation audio on servers, AI providers with zero retention, and no AI training on audio. Free trial is not counted as a free plan, AI rewriting and voice editing commands are not stated, and prices are not dataset rows.

Kept, 6 figuresMacWhisper

Maker's homepage and privacy policy give six features: offline processing, local speech processing, dictation in any app, a free forever plan at 0 euros per month, no data leaving the device, and remote AI models as an option. Custom words, voice editing commands, and dictated-text AI rewriting are not stated; the AI summary and chat features are for transcripts, not rewriting dictation.

Kept, 4 figuresDragon Professional (Nuance)

Nuance's own help and privacy statement give four features: dictation into most Windows applications, 'Scratch/Delete/Undo That' editing commands and vocabulary words in the version 16 help, and a privacy statement that voice snippets may be reviewed to train. Works offline, on-device processing, AI rewriting, and server storage are not stated for the standalone product; NMS mode needs internet, a

Kept, 9 figuresHold to Talk by NS EC Holdings, LLC

Maker's product page and privacy policy (updated October 15, 2026) state offline, device, app-typing, word list, free plan, training, audio retention and OpenAI facts; no statement on voice editing commands

Kept, 9 figuresHyperWhisper by HyperWhisper

Maker's home page, Data Privacy guide and Voice Commands docs state offline local mode, any-app typing, AI modes, vocabulary, voice formatting cues, free tier and cloud LLM use; no statement on training

Kept, 5 figureshyprwhspr by hyprwhspr

Maker's home, docs and privacy pages state offline-by-default, local speech, Linux typing, free MIT license and no audio sent; no AI text rewriting, vocabulary or voice editing stated

Kept, 2 figuresJust Press Record by Open Planet Software

Privacy policy says audio is never uploaded to the maker's servers and that iOS transcription typically happens on the device; the product page and policy say nothing on offline use, typing into other apps, AI rewriting, custom words, voice commands, a free plan or training

Kept, 8 figuresKeyVox by Dominic Esposito

Home page, FAQ and privacy policy state offline use after the model download, on-device transcription, insertion into other Mac apps, Vibes writing styles, a custom dictionary, always-free dictation, and no audio upload or text leaving the Mac; no statement on voice editing commands or on training

Kept, 8 figuresKoe by Koe

Docs, privacy policy, product page and sponsor page state local Apple Speech and MLX options, auto-paste, optional LLM correction, remote LLM text, a plain-text dictionary, no Koe-operated relay and free use; no statement on voice editing commands or on training

Kept, 4 figuresLazyTyper by LazyTyper

Home page and privacy policy give rows for offline local models (with the FAQ's internet requirement for cloud engines noted in the detail), free use with no ads, and no storage of voice recordings; no page sentence on typing into any app, custom vocabulary, voice commands, AI rewriting, or text sent to an AI, so no rows for those; audio goes to a chosen speech API, not text, so no text-to-AI row.

Kept, 8 figuresMacParakeet by MacParakeet

Home page and privacy policy give rows for offline use after the model download, on-device speech on Apple Silicon, typing into any app with a text cursor, AI transcript cleanup, custom vocabulary, a free GPL-3.0 build with no paid tier, audio that never leaves the Mac, and opt-in text to AI providers; no page sentence on voice commands or on recordings training models, so no rows for those.

Kept, 7 figuresMuesli by Muesli

Home page and privacy policy give rows for offline use after install, on-device speech after a model install, pasting into the current app, a personal dictionary, a free open-source build, recordings not sent to its servers by default, and optional meeting summaries sent to an AI provider; the agent CLI docs were read and add no dictation feature, and the policy does not say whether recordings tra

Kept, 9 figuresMurmure by Murmure

The maker's homepage, docs and privacy policy state a free plan, custom dictionary, Command Mode, on-device transcription, no cloud upload by default, and an optional remote LLM.

Kept, 4 figuresnerd-dictation by ideasman42

The project README gives offline VOSK recognition, a local model folder, simulated keystroke output and a grammar file. It states nothing on AI, price, recordings or servers, so those get no row.

Kept, 4 figuresOffline Voice Input by notune

The project README states that speech is transcribed on the device, typed into any app, and that no audio leaves the phone. It names no AI, custom vocabulary, voice commands, price or training use, so those get no row.

Kept, 8 figuresOpenbroca by Openbroca

Home page FAQ says cloud recognition needs an internet connection; privacy policy says audio goes to a third-party AI provider and is not stored by the maker; pricing page gives a free plan of 3,000 units a week with every feature, focused-field insertion, text refinement and a personal dictionary. Not stated: voice commands for editing, and whether recordings train models (the policy covers trans

Kept, 6 figuresOpenLess by OpenLess

README says it bundles a local speech model on macOS, inserts results at the cursor, polishes text with AI, keeps a dictionary of hotwords, says all data stays on the machine, and lists OpenAI and Google Gemini as polish providers. Not stated: whether it works offline (every polish provider listed is a cloud service), a free plan or price, voice commands for editing, and model training on recordin

Kept, 1 figureOpenSuperWhisper by Starmel

README's contribution to-do list puts a custom dictionary and keyword boosting on the list of unfinished work, so no vocabulary feature is shown. Not stated: offline use or on-device speech (it describes downloading Whisper model files, not where recognition runs), typing into other apps, AI rewriting, voice commands, a free plan, and any data handling.

Kept, 8 figuresOpenTypeless by OpenTypeless

Home, pricing, privacy policy and a data-flow guide from the maker give rows for any app, optional AI polish, custom dictionary, free forever, and transcript text sent to the LLM the user configures. No row for offline use, on-device speech, voice editing or training: the pages only call local routes options (a Custom Whisper endpoint, Ollama) and say nothing about training or voice commands. No a

Kept, 7 figuresOSTT by kristoferlund

Home page and guides from kristoferlund give rows for offline use with a downloaded model, on-device speech with local models, paste into any focused app, AI prompt actions, the keyword list, no subscription, and recordings stored locally (0). No row for training or text sent to an AI service: the pages say the AI prompts run but do not name the service that receives the text. No voice-command row

Kept, 11 figuresParaspeech by Burlis Management GmbH

Home, pricing, documentation and privacy policy from Burlis Management GmbH give rows for offline use after download, on-device local models, insertion into many fields only (0), AI rewriting, word replacements, a 7-day trial rather than a free plan (0), cloud rewrite text sent to Groq, and feedback audio kept 30 days when sent. No training row: the policy says feedback is not used to train models

Kept, 7 figuresPithflow by Pithflow

Pithflow's own pricing, home and privacy pages give 7 rows: free plan, types into any app, AI cleanup, dictionary, speech on the maker's servers, audio not kept, and text sent to Groq's AI. No row for offline use, since the pages do not say it either way.

Kept, 5 figuresQuill Flow by Rishabh Raj

Quill Flow's own home, FAQ and privacy pages give 5 rows: free plan, works in any app, AI formatting, personal dictionary, and no audio stored. The pages name the publisher Flow Studios, not Rishabh Raj; the official site is quillflow.app as the brief gives it. No row for offline use, on-device speech, editing commands, training or third-party AI, since the pages do not say.

Kept, 9 figuresRekody by Rekody

Rekody's own home, privacy and docs pages give 9 rows: works offline, speech on the device, types into any app, optional AI cleanup, dictionary, free plan, no training on recordings, no audio servers, and optional text to an LLM. No row for editing commands, since the snippets are text expansion and the pages describe no editing commands.

Kept, 4 figuresSayboard by ElishaAz

GitHub README by the maker: on-device Vosk speech on Android, and no data sent over the internet. No statement on offline use, AI rewriting, vocabulary, voice commands, typing into apps or a price, so those rows are not written.

Kept, 6 figuresSayCopyPaste by Bravely Studios

Product page and privacy policy by the maker: optional on-device mode for phones, clipboard-only output, custom vocabulary, and no audio retained on bravely.dev. Free plan omitted because the page's trial and the privacy policy's free-tier cap conflict. No AI rewriting or voice commands stated.

Kept, 7 figuresSayso by Sayso

Maker's website: on-device speech by default, text at the cursor, opt-in cloud models and AI styles with your own key, pay-what-you-want price, and voice stays on the computer. No statement on offline use, vocabulary or voice commands, so those rows are not written.

Kept, 7 figuresSotto by Sotto

Home page and privacy policy give rows for offline use after a model download, speech processed on the Mac by default, pasting into any app, AI before pasting, custom vocabulary, audio not sent to the maker, and optional cloud AI for text. No row for a free plan (the pricing page lists only a one-time purchase) or for voice commands for editing, and no row for training because the policy does not

Kept, 3 figuresSpeakApp AI by VoicePop Inc.

Home page gives a row for AI rewriting. The privacy policy from Voicepop Inc. gives rows for recordings stored by hosting providers in EU data centers and for AI service providers processing data. No row for offline use, on-device speech, typing into other apps, custom words, voice commands or training: the official pages do not say so. 'Free to try' is a trial, not a free plan, so no free-plan ro

Kept, 4 figuresSpeakly by Genspark

Home page and FAQ give rows for a free plan with 4,000 words per week, typing where the cursor is in any app, and AI auto-edits. The privacy policy from MainFunc Inc. and Genspark Inc. gives a row for queries sent to the OpenAI or Anthropic API. No row for offline use, on-device speech, audio storage or training: the policy does not describe dictation audio, and custom instructions are AI prompts,

Kept, 9 figuresSpeakoFlow by SpeakoFlow

Its home page FAQ, privacy page, and AI cleanup, custom words and dictation guides give rows for offline use, on-device speech, typing into any app, AI cleanup, custom words, spoken edits, a free plan, feedback that excludes recordings, and cloud AI only when the user adds a provider. No sentence says whether recordings train models, so no training row.

Kept, 6 figuresSpeakType by SpeakType

Its official home page gives rows for offline transcription, on-device speech, typing into any app, a free and open-source app with no subscription, zero data leaving the Mac, and nothing leaving the device. It does not mention custom words, voice commands, AI rewriting or model training, so no rows for those.

Kept, 5 figuresSpeech Note by mkiol

The project README gives rows for offline use, local speech processing, typing into any focused window through global shortcuts, and no data sent to the Internet. It does not mention AI rewriting, custom words, voice commands, a free plan or model training, so no rows for those.

Kept, 2 figuresTalon Voice

Official home page shows spoken text-editing commands, and the license agreement grants a royalty-free license. The pages do not say whether speech is processed offline or on the device, so those rows are not written.

Kept, 7 figuresBraina Pro

The official download page lists Pro dictation as offline-capable in 90 languages, with AI correction, custom words and a free Lite edition. The page does not say where the speech is processed, and it names no voice-command editing feature, so those rows are not written.

Kept, 9 figuresVoibe

Voibe's own pages give rows for a free plan with limited words, custom vocabulary, typing into any app, on-device mode that works offline and speech transcribed on the Mac, AI cleanup in Cloud Mode, and no training on or storage of recordings. Rows already staged from other jobs are reused. No Voibe page names voice commands for editing. The Cloud Mode text names Cerebras as the host of its model,

Kept, 3 figuresSpeech Recognition Anywhere by Sea Breeze Computers

Product page by the maker: 30-day free trial and then paid (no free plan), Windows add-on reaches Office and Word, built-in undo command. No statement on offline use, on-device speech or vocabulary; privacy statement covers the website and chatbot, not the extension's audio, so those rows are not written.

Kept, 5 figuresSpeech to Text (Voice Recognition) by Keller

Chrome Web Store listing by the developer: Whisper engine works offline and runs in the browser, captured text stays in the extension's text box, developer says it will not collect or use data. Linked privacy policy describes an autofill add-on with no microphone, so it is not used. No statement on free plan, voice commands, vocabulary or AI.

Kept, 3 figuresSpeechAgent by add0n

Product page by the maker: speech recognition occurs on a server through the browser, the extension has no server-side code, and it types only into web page text areas. Privacy policy covers the website, not the extension's audio. No statement on offline use, AI, vocabulary or voice commands.

Kept, 3 figuresSpeechTexter

SpeechTexter's home page and privacy policy give three yes or no answers: a free web app with no registration (1), voice commands for undo, redo and new paragraphs (1), and speech processed at Google servers, so not on the device (0). No statement on offline use, types into any app, AI rewriting, custom vocabulary, training on recordings, or AI text sharing, so those rows are left out.

Kept, 3 figuresTalk-type

The maker's README gives three yes answers: it writes into the focused field through macOS Accessibility (1), AI cleanup is optional with a provider the user chooses (1), and it lists six AI providers for that cleanup (1). It does not name its speech recognizer as on-device or cloud, offers no free-plan or pricing statement, and says nothing on custom vocabulary, voice commands, offline use or tra

Kept, 3 figuresTalkaType

The support page says recognition needs network communication, so it does not work offline (0). The home page describes it as a browser feature for typing into websites (0 for any app) and offers a free Chrome extension (1). The privacy statement covers personal data but not speech audio, so no server-storage row. Nothing on AI, voice commands, vocabulary or training.

Kept, 4 figuresTalknotes by Talknotes

Home page FAQ and styles give rows for rewrites with AI (1), internet needed for full features (0), audio deleted after transcription (0), and no training on notes (0); pricing and privacy pages give no feature row (free trial is not a free plan).

Kept, 2 figuresTalkTyper by TalkTyper

Home page calls it free speech to text in a browser (1), and the privacy policy says recordings are sent to Google's servers (speech on device 0); no statement on offline use, custom words, commands or AI, and copy and paste is not a statement about typing into apps.

Kept, 7 figuresTambourine Voice by Tambourine

Home page gives rows for local models, any text input, personal dictionary, spoken corrections, AI formatting, free self-hosting and cloud providers; Selection Commands, Voice Shortcuts and the hosted service are marked Coming Soon and are not counted, and offline use is not stated.

Kept, 1 figureTranscribro by soupslurpr

The README says it is a private, on-device speech recognition keyboard for Android, English only. It does not say offline, cloud, AI rewriting, custom words, voice commands, price or data use, so no rows for those.

Kept, 8 figuresTypeflux by Typeflux

The maker's home page and privacy policy state offline use after a local model download, on-device models, any-app typing with the Fn key, AI rewriting and voice commands in Ask Anything, a free core with an optional Pro plan, no collection of voice data, and optional third-party providers. No training row: the privacy policy says processed data is used for product improvement, which conflicts wit

Kept, 7 figuresTypeNo by marswaveai

The maker's home page says all speech recognition runs on the Mac with no network request, pastes at the cursor, is free forever under GPLv3, and that no data leaves the device. It says nothing about custom words, voice commands, AI rewriting or a paid tier, so no rows for those.

Kept, 10 figuresTypeWhisper

TypeWhisper's own home, pricing, privacy policy and macOS documentation give all ten features: offline after model download, local engines, typing into any app, AI workflows, a dictionary, inline spoken commands in dictation-only workflows, a free GPLv3 tier, no audio on its servers by default, audio excluded from opt-in training contributions, and cloud AI only after the user configures a provide

Kept, 9 figuresUtter

Utter's home page, privacy policy and offline guide give nine rows: offline with a downloaded model, on-device transcription with a local Parakeet model, typing into any app, AI Modes, custom replacements, a $0 plan, no training on user content, no audio kept on its servers after processing, and transcript text sent to a provider only when optional AI is on. No row for voice commands for editing:

Kept, 6 figuresVocalinux

Vocalinux's own privacy policy and home page give six rows: offline after the model download, free with no premium tiers, no audio sent to its servers, and no recordings sent to the maker. Two rows reuse catalog facts from the project's GitHub README: on-device speech with local engines, and typing into the focused app. No row for custom words, AI rewriting, voice commands or third-party AI text:

Kept, 3 figuresVoice Control for ChatGPT by Aidia ApS

The home page offers a free extension and says spoken questions are submitted to ChatGPT; the support page says recognition needs network communication. No rows for on-device speech, any-app typing, AI rewriting, custom words, editing commands, training or audio storage, because the pages do not say. Voice Control Pro, by the same team, is a separate app and is not counted.

Kept, 3 figuresVoice Notebook by Voice Notebook

The help page says the service uses Google back-end servers for speech recognition and is not offline; the home page says an integration module types into any application on Windows, Mac and Linux. No row for editing voice commands: the page says voice commands work during dictation but does not say they edit text. No row for custom words: the page mentions a replacement list, not vocabulary. No f

Kept, 3 figuresVoice Typing Notepad by getvoicetyping.com

The home page names 'scratch that' as a spoken editing command and says audio goes to Google's speech service, so speech is not processed on the device; the privacy policy says the developer does not receive, store or have access to the audio. No offline row: the page says dictation relies on Chrome's speech service but does not say whether it works offline. No any-app row: dictation is limited to

Kept, 10 figuresVoicenotes by Voicenotes

The maker's pricing page gives Basic at $0, Pro at $9 and Enterprise at $24 per user per month, a 100-plus language count (its FAQ says 60-plus, so both are recorded), and 30-day history on Basic. The privacy policy says recordings are stored on AWS in the United States, content goes to third-party AI providers, and recordings are not used for training. No row for offline use, typing into any app,

Kept, 11 figuresVoiceOS by VoiceOS

The home page says it works in apps such as Gmail and Slack, removes filler words, fixes grammar and formats text. The pricing page lists custom vocabulary and voice editing of selected text, 100 dictation languages, a 7-day trial with no free plan, and Pro at $11.99 per month billed annually. The privacy policy says speech and prompts go to providers, and that recordings may be kept for training

Kept, 1 figureVoicepal by Voicepal Limited

The home page says Voicepal's AI turns spoken thoughts into first drafts, trained on the user's tone and writing structure: one row for AI rewriting, with the page's own wording. The privacy policy is linked to a Notion page that shows no text in a plain fetch, so there is no data row for training, servers, third-party AI or on-device speech. No offline, typing, custom words, voice command or pric

Kept, 7 figuresVoicetypr by Voicetypr

Voicetypr's home page and privacy policy state offline dictation by default, on-device Whisper or Parakeet speech with a local model, pasting into any app, custom vocabulary, and optional AI cleanup with the user's own key. The privacy policy says the hosted API does not receive audio in local mode and that the configured AI provider gets the text. No row for a free plan (the 3-day free trial is n

Kept, 10 figuresVoiceVibe by VoiceVibe

VoiceVibe's pages state a Free Forever tier with on-device voice typing that works offline, typing into any text field, Pro AI text transformation, a Pro personal dictionary, spoken edit commands, and Voice Assistant context sent to AI providers. The privacy policy says it does not use data to train AI models. The audio row is 1 because voice clone audio is kept until deleted, and the detail says

Kept, 7 figuresVoicy by UseVoicy.com

Voicy's home page and privacy policy state offline mode on Mac, Windows and Linux with nothing sent to its servers, dictation into all websites and apps, AI rewriting for emails, and no training on recordings. The privacy policy says recordings and transcripts are not stored and that online mode sends text through Groq. No row for custom words or voice commands: the pages do not say. No row for a

Kept, 6 figuresVOXD by VOXD

VOXD's GitHub README says voice processing is local and offline with Whisper.cpp, types into any Linux app, offers optional AI rewriting with local or cloud LLMs, and is open source and free of charge. The README lists OpenAI, Anthropic and xAI as cloud providers. No row for custom words, voice commands or training: the page does not say. No row for keeps audio on servers: the README says local, b

Kept, 6 figuresVoxtype by Voxtype

Voxtype's FAQ says all speech recognition is done locally with whisper.cpp and that it does not need an internet connection; its FAQ says the text is typed at the cursor position or copied to the clipboard. The home page says the audio stays on the machine and offers an optional post-processing pipe through any LLM. The configuration reference describes an initial prompt for domain vocabulary. No

Kept, 9 figuresWeesper Neon Flow by Weesper

Weesper's home page says 100% offline dictation for Mac and Windows and text at the cursor in supported apps; its privacy policy says speech recognition runs on the device and audio is not sent to its servers for transcription; its pricing page gives euro prices (monthly 5, annual 45 a year, lifetime 99 once) and custom dictionaries. The FAQ says users try all features for 15 days and then choose

Kept, 4 figuresWhisper IME by woheller69

The GitHub readme says voice recognition works entirely offline once the Whisper models are downloaded, and that it works as a keyboard (IME) and as system-wide voice input. Rows: works offline, speech on device, types into any app, and no audio on the maker's servers (0, read from 'entirely offline'). No row for AI rewriting, custom words, voice commands, a free plan (the readme gives only the MI

Kept, 3 figuresWhisper Memos by Median Tech, s.r.o.

The privacy policy, dated October 1, 2026, says transcription sends each audio file to OpenAI (speech not on the device, 0) and that transcribed text goes to the GPT-4 API for paragraphs (AI formatting and third-party AI, both 1). No row for offline use, custom words, voice commands, training on recordings, or audio kept on the maker's servers: the pages do not say. No row for a free plan: the pri

Kept, 2 figuresWhisper To Input by j3soon

The GitHub readme says the app works in any app that takes text input, such as a browser (1). It says recognition runs on the OpenAI API or on a Whisper server or NVIDIA NIM backend the user sets up, so speech is not processed on the device (0, reused from the existing fact on the two backend choices). No row for works offline: the readme does not say whether the app runs without a backend. No row

Not keptSolace Dictate

No official page found. Two web searches for 'Solace Dictate' return only unrelated apps named Solace and a different recorder (Dictate + Connect), with no maker page for a dictation product.

Kept, 6 figuresBetterDictation

The maker's home page (betterdictation.com) gives six rows: basic dictation works fully offline, speech is run by Whisper on the Neural Engine, it types into any app, Pro adds an AI pass to every dictation, Pro sends transcripts to OpenAI, and the software never stores audio. The page does not mention custom words, voice commands for editing, or a free plan (only free trials), so those get no row.

Kept, 5 figuresApple Dictation on Mac

Apple's Mac Help and feature availability pages give five rows, all reused from existing facts: on-device processing of general Dictation, the Monterey 12 no-internet statement (no current Mac Help line says offline; the row's detail names the OS), typing anywhere, voice editing through Voice Control, and audio stored only if the Share Audio Recordings opt-in is chosen. No row for custom words, si

Kept, 9 figuresWhisperTyping by WhisperTyping

The home, pricing, privacy and FAQ pages state nine rows: not offline (FAQ), cloud speech and no audio stored (privacy), typing into any app, AI Rewrite mode, custom vocabulary on the free plan, voice rewrite of selected text, a free plan, and text sent to OpenAI for AI modes. No row for training on recordings or for a fact about the maker's servers beyond storage: the pages do not say.

Kept, 3 figuresWhisperWriter by savbell

The GitHub README states three rows: speech transcribed by a local faster-whisper model by default with the OpenAI API optional, text written to the active window, and GPT post-processing listed as roadmap (0). No row for offline use, a free plan, custom words (only an initial_prompt option is described, and word replacement is roadmap), voice commands, training on recordings, or sending text to a

Kept, 6 figuresYap by Frigade by Frigade

The product page and the maker's GitHub README state five rows: works offline once macOS has the speech model, on-device Apple Speech transcription, pasting into any app, free to use, and audio that stays on the Mac. The README adds no network calls, so no text goes to a third-party AI. No row for custom vocabulary, voice commands, AI rewriting or training on recordings: neither page says these.

Kept, 7 figuresWindows voice typing (Win+H)

Microsoft's help page gives seven rows: needs internet, online Azure recognition, any text box, fluid dictation on Copilot+ PCs, delete and select commands, opt-in voice clips for training and kept up to two years. Win+H itself is on the page; no free plan or custom vocabulary is stated.

Kept, 2 figuresGoogle Docs voice typing

Google's Docs Editors Help page gives two rows: voice typing works only in Google Docs and Slides speaker notes and captions, and editing commands exist (English only). The page does not say where speech is processed, whether it works offline, or whether there is a free plan, so those rows are left out.

Kept, 7 figuresGboard voice typing

Gboard's Help pages and App Store listing give seven rows: types into most apps, speech stays on device for advanced voice typing (Pixel 6 and up), detailed-edit voice commands (Pixel 8 and up), writing tools (Pixel 9 and up), opt-in speech snippets for training and kept up to 18 months, and a free app. The iPhone listing says voice input is sent to Google for transcription, so the on-device row i

Kept, 5 figuresMicrosoft Word Dictate

Microsoft's Dictate pages say it needs a microphone and a reliable internet connection, that speech is sent to Microsoft, that it is only for Microsoft 365 subscribers, and that the service does not store audio; the English command table lists undo and delete that. No official page gives a figure for AI rewriting, custom words or training, so those cells are left out.

Kept, 5 figuresOtter.ai

Otter's own pricing, privacy policy and security pages give rows for free plan, custom vocabulary by plan, AI training on de-identified audio, AI service providers and AWS S3 storage. No Otter page we could reach says whether it works offline or processes speech on the device, and the only statements are third-party reviews, so those cells are left out. Its help center (support.otter.ai) could not

Kept, 9 figuresLetterly

Letterly's help center and home page give rows for offline use (recording only), cloud transcription, dictation in any app, AI rewriting, a Dictionary, free plan limits, no AI training, storage on Google Cloud and third-party providers. Voice commands for editing are not described on any page, so that cell is left out.

Kept, 8 figuresAudioPen

AudioPen's own pricing, FAQ and privacy policy give rows for a free plan, typing into any app, AI rewriting, an offline option in its apps, local mode, no AI training, 48-hour deletion of voice-note audio and third-party AI providers. No AudioPen page says it has custom words or voice editing commands, so those cells are left out.

Kept, 9 figuresTypeless

Typeless's own home, pricing and privacy pages give rows for a free plan, typing into any app, AI rewriting, a personal dictionary, speak-to-edit commands, no training on dictation, no stored audio, third-party AI providers and cloud processing of speech. No Typeless page says whether it works offline, so that cell is left out.

Kept, 8 figuresHandy (open source)

Handy's README, project page and privacy policy give rows for offline use, on-device speech, typing into any text field, optional AI post-processing that runs only when invoked, custom words kept on the device, a free app and no audio sent to Handy's servers. No Handy page says whether it uses recordings for training or has voice editing commands, so those cells are left out.

Kept, 9 figuresOpenWhispr (open source)

The maker's README, cleanup docs, privacy policy and product page give a yes or no for all but one feature: offline local models, on-device speech in local mode, cloud audio not stored, no training on audio, language-model cleanup with cloud or own-key providers, custom dictionary, any-app typing and a free start. Voice commands for editing are left out because the page says only basic commands.

Kept, 7 figuresWhispering (open source)

The current build's README (Epicenter repository) gives a yes or no for most features: no on-device transcription, no dictation into other apps, recordings local, transform with prompts and a chosen text endpoint. The free row comes from the archived repository's tagline, which describes the earlier app. Custom words, voice commands and training are not stated.

Kept, 4 figuresVoice In (Chrome extension)

The developer's website and privacy policy give a yes or no for voice commands for editing, the free plan, audio and transcripts not sent to the developer's servers, and no external transmission of text. Offline, on-device speech, any-app typing, AI rewriting and custom words are not stated: the browser's built-in speech engine does the transcription, and neither page says where it runs.

Kept, 3 figuresFUTO Voice Input

FUTO's product page says the Android app is designed to work entirely on-device, and its privacy policy says microphone data is always processed on the device and never transmitted elsewhere. These give rows for works offline, speech on the device, and no audio on the maker's servers. No FUTO page mentions AI rewriting, custom words, voice commands, a free plan or AI training, so those cells are l

Kept, 3 figuresMicrosoft SwiftKey voice input

Microsoft's SwiftKey voice typing help article says SwiftKey may send anonymized voice clips to Microsoft to improve recognition, with an opt-out in SwiftKey settings, and that offline recognition is set up in the Google app for the Google Voice Typing mode. The offline row is limited to that mode, and the default mode is not described as offline. The article does not say whether voice clips are k

Kept, 1 figureSamsung voice input

Samsung's US keyboard support page has an Offline voice input section where you install offline language packs, which gives the works-offline row (reused fact). The page does not say speech is processed on the device, so that cell is left out. Samsung's US privacy policy keeps voice recordings on its servers for the voice command feature, not for voice input, so no server row is written. No page g